DeepSeek V4.1 Flash يتوفر رسميًا للعموم: 552B بمعمارية جديدة، أوزان MIT، وإيقاف V4 Pro الذي لم يحدث قط

DeepSeek V4.1 Flash is no longer a two-day experiment. The model that spent September 8 and 9 sitting inside the vendor's live API under an ID carrying its own expiry date — deepseek-v4.1-flash-expires-on-0910 — was released for real on September 10, 2026, across DeepSeek's app, web interface and API, where it now answers to the model name deepseek-flash. Three things about the release are larger than the beta implied. The first is the paperwork: DeepSeek published a technical report — titled, without much subtlety, Pushing the Limits of KV Cache Compression — and MIT-licensed weights for a 552B-parameter multimodal mixture-of-experts model built on a new Causal Encoder-Decoder architecture. Those weights are not a promise: 48 safetensors shards, about 510 GB in total, went online the morning of release, which turns the beta's unverified “new architecture, native multimodal” slogan into something a reader can download and inspect. The second is the shape of the line-up. DeepSeek V4 Flash and DeepSeek-V4-Flash-Vision-Exp are retired, their old IDs still resolving to the new model at Flash rates, and the plan to phase DeepSeek V4 Pro out on September 14 was withdrawn before the date arrived: V4 Pro is still served, at unchanged billing, and DeepSeek has promised notice before anything changes. The cheapest model in the family has taken the flagship's place on price and on DeepSeek's own benchmarks, and DeepSeek says the architecture underneath it was designed to scale to larger models. What sits above it is, as of September 25, one tracker's guess — a three-tier Flash, Pro and Ultra line at 522B, 2T and 8T — offered by the account that posted it as speculation and confirmed by nothing DeepSeek has published. The third is the one the beta never had. Artificial Analysis has now published the first outside evaluations of the model — an Intelligence Index score of 39 in Reasoning mode at maximum effort for the September weights and 45 for the 1017 checkpoint (a six-point gain that ties it with GLM-5.3 for second among open-weights models, a point behind MiMo-V2.6-Pro), against 36 for DeepSeek V4 Pro 0813 on the same scale, which makes the 552B model the highest-scoring DeepSeek model the site has measured. Ten further items are younger than the release, and they point in different directions. On September 25 ARC Prize's own model-testing lead opened a pull request adding a first-class DeepSeek V4.1 Flash profile to the foundation's benchmarking harness — the evaluator whose board this field reads as a reasoning ceiling rather than a productivity check, arriving with no result attached and none promised. Three days later an outside lab published the opposite kind of news: a finished measurement, on a different board, that puts DeepSeek V4.1 Flash next to another open-weight model and finds it the weaker of the two — and finds the number moving more with the harness wrapped around the model than with the model itself. On Exploit Bench, a Carnegie Mellon benchmark scoring how far an agent climbs toward full exploitation of a production V8, DeepSeek V4.1 Flash averaged 4.7 of 16 under the benchmark's own default configuration and 9.3 under Claude Code, while GLM-5.3-Flash peaked at 10.6 under Codex and fell to 6.2 under the very harness that suits DeepSeek V4.1 Flash best. The third is the least dramatic and the easiest to check: on October 7 the vLLM project published its own three-week accounting of serving this model, reporting its current build 1.9× faster at low concurrency and 5.3× the throughput under a 150 TPS constraint against its own day-0 implementation — an engine's before-and-after on itself rather than another outside evaluation, and the subject of the speed section below. A fourth is about what the model costs to use rather than what it scores, and it moved twice in three days: on October 12 OpenCode, the coding agent that prices this model by monthly usage allowance rather than by metered tokens, raised its allowance for it from $60 to $80 — the platform's own statement about a product it sells, carried on its own pricing page — and on October 15 it began serving the same model free, under a second model id, on a 200,000-token context against the 1M window the metered entry carries. A fifth, from October 14, measures the model's cost to serve rather than its quality: Artificial Analysis added DeepSeek V4.1 Flash to AA-AgentPerf, its hardware benchmark for agentic inference, with results on both NVIDIA and AMD accelerators — at 100 tokens per second per user, NVIDIA's GB300 NVL72 leads the field at roughly 261 million tokens per second per megawatt, and the same runs serve the model on as few as two accelerators, which is the section below in a second currency. The sixth is the release itself, superseded: on October 17 DeepSeek replaced the checkpoint behind the same deepseek-flash model name with DeepSeek-V4.1-Flash-1017 — new weights, a vendor benchmark table that beats the September model on every row it publishes, the same rate card and the same 1M-token envelope — and, as of October 18, an outside measurement of the new weights that contradicts the row the vendor leads with: Artificial Analysis scores the 1017 checkpoint 45 on its Intelligence Index and 33.8 on Terminal-Bench 4.0 under its own harness, against the 42.3 on DeepSeek's card. The seventh item is the argument that gap produced. On October 19 a tracker account with a record of calling these releases early declared itself “dead certain” that the 42.3 is real “because it's better than GLM-5.3,” and asked Artificial Analysis to run the model in a native harness. The site had already answered the day before, with a number eight and a half points lower. The eighth is not a score: on October 20 a DeepSeek tracker posted an agent transcript in which the 1017 checkpoint declined a jailbroken persona with dsh-purge — a plugin built to strip the harness's own constraints — applied, and a third party's abliteration notes put the untouched checkpoint's refusal rate at 97 of 100 harmful prompts in chat mode and 98 in think mode, with DeepSeek publishing no safety evaluation of its own for either checkpoint. The ninth is not a score and not a price: this week DeepSeek opened the inference engine its own service runs on, with a decode curve for the 1017 checkpoint and the production figures behind it, and the thread that surfaced it this week lays that curve against a September 30 Ascend 950DT benchmark that puts Huawei's part roughly five to seven times behind per chip — the serving-silicon section below. The tenth is not a score, a price or a checkpoint: on October 22 YuyuantanTan, the current-affairs account run by China Media Group, reported that DeepSeek and MiniMax had completed the registration procedures attached to China's draft technology-export catalogue and finished open-sourcing and overseas licensing. That is state media reporting on two companies rather than a regulator's notice or a company statement, and it is the development that answers the question the draft catalogue opened on October 16 — whether the MIT weights this piece is about can keep being published and shipped abroad. The licence section below separates what the report says from what it does not.

Everything below keeps two kinds of claim apart. The specifications, benchmarks and prices come from DeepSeek's own changelog, model card and technical report — vendor-published, vendor-run, and not reproduced by anyone else. That gap has since narrowed by four independent instruments — a composite index, a community kernel board, a harness-controlled cyber benchmark and a hardware serving benchmark — one of which, the index, has now been run against both checkpoints. None of them replicates a single vendor row, and ARC Prize is still building the reasoning evaluation, with nothing published from it yet. DeepSeek replaced the model behind the same API name on October 17 with the DeepSeek-V4.1-Flash-1017 checkpoint. For a day that meant every outside number in this piece described weights the vendor no longer served, with no external score for the ones it did; on October 18 Artificial Analysis re-ran its index on the new checkpoint and published 45, against 39 for the September weights. The other three instruments have not been re-run: the cyber benchmark, the kernel board and the hardware benchmark still describe the September weights, and no independent score, old or new, replicates one of DeepSeek's own benchmark rows. The cost-of-serving figure the vLLM project published on October 17 is newer only in publication date: the chart behind it is dated October 16, the day before the swap, so it describes the same superseded weights. On its current index (v4.3.2) Artificial Analysis scores the September DeepSeek V4.1 Flash checkpoint at 39 and the 1017 checkpoint at 45, describing both as well above the median of 19 for open-weight models of comparable size; they are genuine third-party numbers and the first in this story that DeepSeek did not produce. Each is one composite of ten evaluations, not a replication of any individual vendor row, and it says nothing about the 1M-context economics. The refusal figures in this piece are a fourth kind of claim and are labelled where they appear: they come from people who build uncensored derivatives of these weights, not from DeepSeek and not from an evaluation lab, and DeepSeek has published no safety evaluation for either checkpoint. The hardware numbers in the serving section are a further class again: the NVIDIA decode curve and the production figures are DeepSeek's own measurements of its own engine, published with the code this week, and the Ascend points come from a benchmark sheet reproduced in a thread on X that is not published on either vendor's site. Where a number belongs to DeepSeek, this piece says so. One more class sits outside that taxonomy. The October 22 report that DeepSeek and MiniMax completed registration under the draft export catalogue is state media — China Media Group's YuyuantanTan, repeated across Chinese financial outlets — not a Ministry of Commerce notice, and DeepSeek has published no statement confirming it. The catalogue itself is a government draft open for comment until November 15, and is described as a draft wherever it appears here. The build names attached to the report — V4.1F-1017 and M3.1-Flash — come from a single X account on October 23 and do not appear in the state-media report or the outlets that relayed it; the checkpoint that post names had already been downloadable for six days when it was written.

The October 17 checkpoint: what DeepSeek-V4.1-Flash-1017 changed

Five weeks after the release this piece is built around, DeepSeek replaced the model underneath it. The changelog entry dated October 17 is three sentences long: “DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017. The official web and APP models have also been updated to the new version. The API calling method and pricing remain unchanged — simply use deepseek-flash to access the latest model.” Everything in that quote is the vendor speaking about its own product, and all three claims are checkable on DeepSeek's own pages rather than taken on trust: the model-version row on the Models & Pricing table now reads DeepSeek-V4.1-Flash-1017 where it read DeepSeek-V4.1-Flash, the rate card beside it is the same off-peak and peak split published on September 10 — $0.003, $0.15 and $0.60 per 1M tokens off-peak, doubling to $0.006, $0.30 and $1.20 inside the weekday peak windows — and the API quick start still names deepseek-flash as the model string, with the 1M-token context, 384K output ceiling and 2,500-request concurrency limit unchanged. The retired alias IDs behave the way the September note said they would: deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted and now answer as DeepSeek-V4.1-Flash-1017 at Flash prices. DeepSeek announced the same change from its own account that day, and the post is worth reading alongside the changelog because it is shorter and blunter: “DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API,” followed by the same three harnesses and the same instruction to upgrade to DeepSeek Harness 0.2.1-rc.1 or later. Everything in it is the vendor describing its own product on its own channel: “now live on our website, app, and API” is the changelog's “the official web and APP models have also been updated” restated, and the benchmark, runtime, output-length and Pareto claims are the model card's — with one difference that matters for anyone deciding whether to try it. The card carries a number next to every benchmark row; the post carries none, and neither the post nor the changelog puts a figure on the runtime and output-length claim. Two of the four things DeepSeek is leading with here are still vendor assertions rather than published results, and the chart behind the Pareto line — the one attached to the post — is DeepSeek's own, drawn on its own task-cost measurements.

The weights moved with it, and that is the part a reader can verify without trusting a changelog. A second repository, deepseek-ai/DeepSeek-V4.1-Flash-1017, went up on Hugging Face in the early hours of October 17 under the same MIT licence: 48 safetensors shards again, the same total of about 510 GB, every shard the same size as its September counterpart, and every shard's hash different. Same shape, different contents — which is what “a minor update” should look like from the outside, and which also means the architecture claims this piece has been reporting since September are untouched: the config file that declares the 552B backbone, the 384 routed experts, the 8B-active-parameters-per-token prefill path and the 16B decode path is byte-identical to the one in the September repository, so the 890-byte-per-token KV cache is the same design rather than a new one. What the new repository does not carry is the paperwork. There is no technical report in it, no evaluation folder and no Jinja chat template; the model card says the release does not include a chat template and points instead at a Python encoding reference and a new Rust toolkit, deepseek-recipe, for turning Chat Completions, Responses and Anthropic-format requests into the model's prompt format. Anyone pinning weights by repository name should know the two downloads differ in more than their hashes. The file list says exactly how: all 48 shards carry a different checksum and an identical byte size, while the files that are not weights — the config that declares the architecture, the licence and the whole encoding folder — are byte-for-byte the same in both repositories. Seven files exist in September's repository and not in this one, and nothing was added to replace them: the technical report, the evaluation folder, the chat template, the two figures the card used to embed, and a .gitignore. The model card narrows one serving parameter as well. September's card recommended either a top_p of 0.95 or 1.0; the October 17 card recommends 1.0, with temperature at 1.0, a 1M context window and a max_tokens of at least 256K unchanged.

DeepSeek's own benchmark table is where the update is meant to show, and it is a vendor table of exactly the kind the September rows were: same-company runs, unreproduced by anyone, published on a model whose weights had been public for a few hours. Against the September checkpoint it is up on every row the card carries — HLE without tools 42.6 against 36.8, or 44.0 against 39.1 on the text-only subset; Terminal-Bench 4.0 42.3 against 31.2; DeepSWE v1.1 77.0 against 74.2; ProgramBench 29.1 against 20.3; NL2Repo-Bench 68.2 against 65.4; Automation-Bench v1.0.6 60.5 against 54.8; Agents' Last Exam 35.4 against 31.8; Apex Bench v1.1 72.1 against 64.6; Chartography with tools 83.0 against 78.9; BabyVision with tools 92.5 against 89.6; and ZeroBench-main with tools 51.0 against 49.0, at Pass@5. The same table places 1017 against the frontier models DeepSeek benchmarks against: it now clears GPT-5.6 Sol on Terminal-Bench 4.0, 42.3 against 39.9, while still trailing Opus 5 on 51.8, and it leads both on DeepSWE v1.1, 77.0 against 73.0 and 74.0 — a row read off the chart attached to the vendor's own announcement, and the single most quoted number of the week. On NL2Repo-Bench it moves to 68.2 and still sits behind Opus 5 on 75.3, which is the shape the September table had too: strongest on harnessed agentic work, behind on the rows that reward reasoning depth. The Terminal-Bench 4.0 row is the one to read twice, because it is now the only row in that table with an independent counterpart. DeepSeek ran it under the Claude Code harness; Artificial Analysis runs the same 66 tasks under mini-swe-agent, and its October 18 result for these weights is 33.8 — eight and a half points below the vendor's 42.3. On DeepSeek's own table that row clears GLM-5.3 by 4.4 points, 42.3 against 37.9. On Artificial Analysis's harness the same weights trail GLM-5.3 by eight, 33.8 against 41.9.

One line of small print is worth more than several of those deltas. The September card evaluated Terminal-Bench 4.0 with DeepSeek Harness in its Minimal mode; the 1017 card says it used the Claude Code harness for Terminal-Bench 4.0 and Mini-SWE-V2 for DeepSWE v1.1. The harness study further down this piece measured seven points of spread on one model across seven agent CLIs, so an eleven-point Terminal-Bench 4.0 move that arrives with a different harness under it cannot be read as eleven points of model. The card states which harness it used, which is what makes the row auditable rather than merely impressive, but the row is two changes wearing one number. The independent run separates them, and not in the vendor's favour. Artificial Analysis evaluated the same 1017 weights on Terminal-Bench 4.0 under mini-swe-agent — the harness it holds constant across every model on that board — and got 33.8, against 26.8 for the September checkpoint on the same harness. On a fixed harness the checkpoint is worth seven points, not eleven, and DeepSeek's headline number sits eight and a half points above what the same weights score when the harness is held constant across models. The rest of DeepSeek's framing is about cost rather than score, and it is the same axis the September release was sold on: the changelog claims the update reduces task runtime and output length, calls the result “a new performance–cost Pareto optimum across multiple harnesses,” and says it optimised for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, recommending DeepSeek Harness 0.2.1-rc.1 or later. Runtime and output length are precisely the quantities the verbosity evidence in this piece tracks from the outside, and DeepSeek now claims to have cut both — with no figure attached. One outside measurement of the new checkpoint now exists, and it does not show the runtime or the output cut: Artificial Analysis's index run has the 1017 weights at 288.8 seconds per task against 289.9 for September, and 89,495 output tokens per task against 88,574 — flat on both, with the cost saving landing somewhere else entirely. The post DeepSeek published on its own account the same day compresses all of that to “faster task completion, and shorter outputs,” then calls the pair a Pareto improvement in performance and cost — the changelog's claim in shorter form, from the same source, and just as unmeasured. It is the vendor restating itself, not a second source — which matters, because the number off that chart is the one the outside world has spent the week repeating. The one artefact in the announcement that does carry numbers is the chart attached to the post, and it is DeepSWE v1.1 Pass@1 plotted against average cost per task under a mini_swe_v2 harness. Its caption is worth reading before the Pareto line is: the public leaderboard data underneath it is dated September 22 on the chart itself, three and a half weeks before the checkpoint the chart is being used to promote, and the comparison lines for OpenAI's models are credited to OpenAI's own DeepSWE chart rather than to a shared run. One point on that plot is new. Everything it is measured against was published before it existed.

في اليوم نفسه الذي جرى فيه التبديل، جاء أكثر تفسير جرى تداوله لذلك الجدول من شخص له مصلحة في مكان تشغيل النموذج. نشر يوتشن جين، الذي يعمل على أنظمة الذكاء الاصطناعي في Databricks، أن DeepSeek-V4.1-Flash-1017 «يسجل 77 على DeepSWE v1.1»، وأنه «أداء بمستوى Opus 5 بتكلفة أقل 83 مرة»، وأننا «سنطلقه على Databricks». الرقم 77 هو صف DeepSeek نفسه، مأخوذ من الرسم البياني المرفق بإعلان المورّد، ولم يعده أي طرف خارج الشركة. نقطتا المقارنة هما من جين، وهما تنقسمان بوضوح إلى واحدة تصمد أمام تحقق خارجي وأخرى لا تصمد. عبارة الأداء بمستوى Opus 5 تصمد: اللوحات المستقلة التي تنشر DeepSWE v1.1 تضع اسم هذا النموذج في ذلك النطاق في تشغيلاتها الخاصة. لوحة APEX لدى Mercor — 113 مهمة أصلية عبر 91 مستودعًا، قُيّمت ضمن إطار mini-swe-agent — تدرج Opus 5 Max عند 71.4% ومدخل DeepSeek V4.1 Flash Max عند 71.7%، كما تضع لوحة llm-stats للاختبار نفسه Claude Opus 5 عند 68.8% مقابل 74.2% لـ DeepSeek-V4.1-Flash. القراءة الأمينة للزوج أضيق مما يصوره المنشور: لوحتان تقيسان هذا الاختبار بشكل مستقل تتفقان على أن النموذج على الأقل ندّ لـ Opus 5 عليه، ولا تُظهر أي من اللوحتين 77، لأن كلتيهما تحمل إما رقم سبتمبر أو تشغيلها الخاص، ولا تسمي أي منهما مدخلًا بنقطة التحقق 1017. عبارة الأداء بمستوى Opus 5 قابلة للتأييد كبيان عن نطاق. أما الرقم الذي أنتجها فليس قابلًا للتكرار بعد.

الجزء الخاص بمضاعف 83× من المنشور هو نتيجة حسابه لا رقم منشور، وهو الجزء الذي ينبغي التعامل معه بحذر. لا تعرض أي لوحة نتائج تكلفة لكل مهمة لـ 1017 checkpoint، والمخطط الخاص بالمورّد الكامن وراء ادعاء Pareto الذي يقدمه DeepSeek نفسه يرسم التكلفة لكل مهمة على محور لوغاريتمي يمتد من عشرة سنتات إلى عشرين دولارًا — وهو نوع الفارق الذي يتطلبه مضاعف بهذا الحجم. وبالاعتماد على أسعار القائمة وحدها، تكون الفجوة أضيق من 83×: إذ تتقاضى Claude Opus 5 مبلغ 5 دولارات لكل مليون رمز إدخال و25 دولارًا لكل مليون رمز إخراج، مقابل 0.30 دولار و1.20 دولار لـ DeepSeek-V4.1-Flash داخل نوافذ الذروة في أيام الأسبوع، أي نحو 17× للإدخال و21× للإخراج، وتتغير النتيجة إلى حد كبير بحسب نسبة إصابة الذاكرة المؤقتة وبحسب عدد الرموز التي تستهلكها المهمة. وليس اتجاه حجة التكلفة محل نزاع، هنا أو في أي مكان آخر في هذا المقال. أما المضاعف المحدد فهو ادعاء بشأن مقارنة لم ينشرها أحد.

Databricks هو الجزء من ذلك المنشور الذي يمكن للقارئ التحقق منه في الوثائق، وتقول الوثائق أكثر مما يقول المنشور. فـ DeepSeek V4.1 Flash متاح على Databricks Unity Gateway منذ 11 سبتمبر، أي في اليوم التالي لإطلاق DeepSeek، مع إدخال النصوص والصور عبر Foundation Model APIs الخاصة بالمنصة تحت اسم الخدمة system.ai.deepseek-v4-1-flash؛ وفي 7 أكتوبر، أضافت المنصة له الدفع لكل رمز بأولوية والإنتاجية المخصصة المحجوزة. وهو أيضاً الوجهة المعيّنة للمنصة لهذه العائلة: فقد جدولت Databricks إحالة DeepSeek V4 Pro (0813) إلى التقاعد في 30 أكتوبر، وDeepSeek V4 Flash (0731) في 5 نوفمبر، مع تسمية DeepSeek V4.1 Flash كبديل موصى به في كلتا الملاحظتين. ما لا تقوله أي صفحة من Databricks هو أن نقطة التحقق الخاصة بـ17 أكتوبر قادمة. هذا الادعاء هو موظف واحد يتحدث باسمه الشخصي — إشارة داخلية إلى النية وليس موعداً للإتاحة الفعلية، ولا تزال صفحة النماذج المدعومة تعرض النموذج دون تسمية نقطة تحقق. لذا فإن تغيير الإتاحة هذا الأسبوع مستقبلي، مصدره جهة داخل المنصة وليس المنصة نفسها؛ وما يمكن لعميل Databricks استدعاؤه اليوم هو أوزان سبتمبر، والتبديل الذي كان هذا المقال يصفه — أوزان جديدة تحت الاسم نفسه وبطاقة الأسعار نفسها — هو الشكل الذي سيتخذه ذلك الاستبدال عندما يحدث.

So what did not change is as important as what did. Three of the four independent instruments still cover the September weights and only those: the Exploit Bench harness table, the community kernel board and the AA-AgentPerf hardware runs were all produced before the swap, and the two DeepSWE v1.1 boards that carry this model's name publish an entry with no checkpoint on it. The fourth changed on October 18: Artificial Analysis added a separate row for the 1017 checkpoint, scored it 45 on its Intelligence Index and 33.8 on Terminal-Bench 4.0, and left the September row standing at 39 and 26.8. That is the first time a number in this piece was produced by someone other than DeepSeek against the weights DeepSeek is currently serving. The API change is a checkpoint refresh behind a name DeepSeek had already documented as pointing at the latest Flash model — the opposite of the September plan to make deepseek-v4-pro answer as a different model, which is the distinction developers were drawing when they objected to that one. For a caller, the practical effect is narrow and concrete: asking for deepseek-flash now gets different weights at the same price, with a better vendor-reported table behind them and, since October 18, one third party scoring them — lower on the one row where both exist. On OrcaRouter the model id and the rate card are unchanged too — DeepSeek V4.1 Flash is routed at DeepSeek's own list price with 0% markup passed through — which is what a vendor refreshing a model without moving its name or its price should look like from the caller's side of the line.

ما الذي تم إطلاقه فعليًا في 10 سبتمبر

The changelog entry opens with a single sentence: "Today, we officially release the DeepSeek-V4.1-Flash model." DeepSeek describes it as the smallest member of a new model-structure family, with native visual understanding, and it is the first release in this cycle to arrive with a technical report attached. The API side is a name swap rather than a migration:

• Model name — deepseek-flash on DeepSeek's own API; the base URL, the key and the request format are unchanged.

• مُتقاعِد — DeepSeek V4 Flash وDeepSeek-V4-Flash-Vision-Exp خارج الخدمة. لا تزال معرّفاتهما القديمة مقبولة، وتُوجَّه الآن إلى V4.1 Flash، مع الفوترة بأسعار Flash، لذا تظل التكاملات القائمة تعمل دون أي تغيير في الشيفرة.

• V4 Pro — DeepSeek announced that from 12:00 Beijing time on September 14, deepseek-v4-pro requests would route to V4.1 Flash at V4.1 Flash prices until a V4.1 Pro shipped, then reversed the decision before the date arrived. V4 Pro is still served, at unchanged billing, and DeepSeek says it will give notice of any further change.

• Envelope — نافذة سياق بسعة 1M رمز، وحد أقصى للإخراج 384K، و2,500 طلب متزامن لكل حساب، مع دعم إدخال الصور، ووضع التفكير مُفعَّل افتراضيًا مع خيار إيقافه.

• الأوزان — في اليوم نفسه، ظهر المستودع deepseek-ai/DeepSeek-V4.1-Flash على Hugging Face تحت رخصة MIT: 48 جزءًا من safetensors يبلغ مجموعها نحو 510 GB، والتقرير التقني بصيغة PDF بجانبها، وتنفيذ استدلال مرجعي في مجلد inference.

That last bullet is also where the retired model ID story lands for anyone with deepseek-v4-flash hardcoded in a production path: it still resolves, but what answers is a different model. DeepSeek has been explicit that this is a compatibility shim rather than a promise, which makes it a migration deadline wearing a deprecation notice.

البنية المعمارية، الآن على السجل

قبل يومين، لم يكن هذا القسم ليكون سوى قائمة بأشياء لم ينشرها أحد. التقرير التقني يحسم الأمر، وتبيّن أن الادعاء البنيوي كان حرفيًا لا تسويقيًا. DeepSeek V4.1 Flash هو نموذج خليط خبراء متعدد الوسائط بـ552 مليار معامل، والأجزاء الجديدة هي الأجزاء التي تحدد تكلفة تشغيله:

• العمود الفقري — إجمالي 552 مليار معامل، مزيج خبراء يتضمن خبيرًا مشتركًا واحدًا و384 خبيرًا موجّهًا في كل طبقة، وستة خبراء موجّهين نشطين لكل رمز.

• CED — محوّل من 40 طبقة، مقسوم إلى مُشفّر سببي من 20 طبقة ومُفكّك من 20 طبقة. تُسقَط ذاكرة التخزين المؤقت العامة KV الخاصة بالمُفكّك من الحالات المخفية النهائية للمُشفّر بدلاً من اشتقاقها من الحالات المخفية الخاصة بكل طبقة من طبقات المُفكّك، ومن هنا يأتي معظم التوفير في الذاكرة.

• عدم تماثل التنشيط {{1}}—{{/1}} هناك نحو 8 مليارات معلمة نشطة لكل رمز أثناء التعبئة المسبقة و16 مليارًا أثناء فك الترميز، لذا فإن قراءة مطالبة طويلة أقل تكلفة بكثير من التوليد منها.

• ذاكرة KV المؤقتة — يزعم التقرير أنها 890 بايت لكل توكن من ذاكرة KV المؤقتة العامة، أي نحو ربع DeepSeek V4 Flash، باستخدام تخزين KV الرئيسي بتنسيق FP4 مع مقياس واحد لكل 16 قناة إضافةً إلى إعادة الاستخدام عبر الطبقات.

• الانتباه وفك الترميز — Compressed Sparse Attention 2 مع أوضاع Full وReindex وReuse ومُفهرِس متناثر هرمي، وSingle-Pass mHC، وDSpark لفك الترميز التخميني.

• Engram — جدول ذاكرة شرطية بـ196B معامل، يُتاح الوصول إليه بشكل متفرق عبر بحث قائم على الرموز، ويقع بجانب مسار خليط الخبراء لا داخل مكدس الانتباه. إنه أكبر كتلة أوزان منفردة في الإصدار، والرقم الذي تذكره بطاقة النموذج له هو ما يجعل حجم التنزيل مفهومًا.

• الرؤية — DeepSeek-ViT مع مُسقِط MLP بطبقتين، مُدرَّب منذ بداية التدريب المسبق لنموذج اللغة بدلًا من إضافته إلى قاعدة نصية لاحقًا.

• التدريب — 45 تريليون رمز من مجموعة بيانات متعددة الوسائط، مع انتباه متفرق مُدرَّب بطول تسلسل 64 ألف، وتم تمديد السياق إلى 1 مليون عند علامة 34 تريليون رمز. بعد التدريب، يتم تشغيل SFT، ثم RL، ثم التقطير على السياسة.

• جهد الاستدلال — عدد صحيح قابل للضبط بشكل مستمر من 1 إلى 100، وهو ضبط أدق من المستويات المنخفضة والعالية والقصوى التي يتيحها DeepSeek V4 Pro.

رقم ذاكرة KV المؤقتة هو الرقم الذي له تبعات تتجاوز ورقة المواصفات، وهو الموضع الذي يثبت فيه التصميم جدواه. نافذة سياق بسعة مليون توكن لا تكون مفيدة إلا إذا كنت قادرًا على تحمّل كلفة ملئها. كانت اقتصاديات السياق الطويل لدى الجيل السابق مؤلمة لسببين: فقد كانت الذاكرة المؤقتة العامة تنمو مع كل طبقة تطبّق الانتباه على المطالبة، وكان معظم سعة الذاكرة المؤقتة الدائمة يُنفق على الاحتفاظ بحالة النافذة المنزلقة التي لم تعد مفيدة في غضون دقائق. ويتصدى جواب التقرير لكليهما. أما ما إذا كان المنتج المقدَّم يطابق ذلك عند مليون توكن في ظل تزامن حقيقي، فهو من نوع الادعاءات التي لا تزال تحتاج طرفًا ثالثًا — فالدرجات المستقلة الموجودة الآن تقيس النموذج، لا اقتصاديات السياق الطويل — لكن الآلية أصبحت موثّقة بدلًا من أن تكون مستنتجة، والقسم أدناه يفككها.

داخل ذاكرة التخزين المؤقت: كيف يتحقق فعليًا 890 بايت لكل رمز

The report's title is the thesis. Read it and the efficiency claim stops looking like a side effect of the mixture-of-experts backbone and starts looking like the design brief — which is the reading independent write-ups have converged on since release. The reason it matters commercially rather than academically is where an agent's bill actually lands: DeepSeek's framing is that the cache-hit line is the one that dominates it on long agentic workloads, which makes the size of the cache a first-order cost lever. DeepSeek's numbers put the global cache at 890 bytes per token, roughly 437× smaller than the original DeepSeek V1 and about a quarter of DeepSeek V4 Flash. Run the arithmetic on a full window and it is the difference between roughly 890 MB of global KV for a single 1M-token sequence and about 3.5 GB at V4 Flash's per-token rate. Three changes do most of the compressing: one on how much is stored, one on what gets read, and one on what is kept at all.

The storage format is the most legible of them. The main global cache is held in FP4 using the E2M1 layout with one E4M3 scale per 16 channels — half the width of the FP8 cache DeepSeek V4 stored and a quarter of FP16. What makes that survivable is that DeepSeek quantized the cache during training rather than at serving time, so the noise the narrower format introduces is something the model learned to work around instead of something inflicted on it afterwards. Not everything was compressed that way: the local sliding-window cache stays in FP8, because the report treats it as the part that is sensitive to further narrowing.

Compressed Sparse Attention 2 handles what gets read. Each attention layer is statically assigned one of three modes — Full, Reindex or Reuse — so decoder layers can share key state and indexer state instead of each keeping its own, and the first Full-mode layer builds a candidate pool that later indexing layers are restricted to. Report figures put that pool at 16,384 positions with the sparse selection over the top 512 entries, which is what decouples indexing cost from context length: DeepSeek's claim is that moving from 4K to 1M context adds only around a quarter more compute per generated token, rather than the roughly window-proportional growth the earlier design implied. That single figure is the most useful thing in the report for anyone trying to price a long-context workload, and it is also the one with no outside confirmation at all — none of the outside measurements in this piece has tested the far end of the context window.

The change doing the most work on the persistent side has the least promising name. DeepSeek V4 kept two things in its persistent cache for more than 72 hours under typical workloads: the global KV and the sliding-window attention state. Sliding-window state matters for minutes, but it was consuming close to half of the persistent cache capacity — storage sized for days spent on data with a minute of usefulness. V4.1 Flash drops SWA state out of the persistent cache entirely and parks it in a pool assembled from about 10% of each machine's host DRAM with a minute-level time to live. When a session resumes and that state is gone, SWA Bounded Replay reconstructs it approximately: rather than replaying layers × window tokens, it replays only the most recent window, reported at 128 tokens, and accepts that the rebuilt state is not mathematically identical to a full replay. DeepSeek says it observed little quality impact from the approximation and simulated the same behaviour during post-training. Put that together with the narrower global cache and the persistent KV footprint falls to roughly one-eighth of DeepSeek V4 Flash under identical workloads — a bigger cut than the one-quarter figure that leads the spec sheet, because it is two changes stacked rather than one format change.

Two limits sit on all of it. The report offers no end-to-end latency or cost comparison, so an eight-times-smaller persistent cache is not a demonstrated eight-times-cheaper workload: compression relieves memory pressure, and only the memory-bound part of a bill responds to that. And the saving is not something that happens to you. An engine has to implement the encoder-decoder projection, the three CSA2 modes, the FP4 cache layout and Bounded Replay before the number means anything, which is why 890 bytes is best read as a first-party specification for DeepSeek's own served API rather than a promise about a local build — the quantized conversions for llama.cpp, Ollama and LM Studio in the repository do not carry this cache with them. The headline also covers text tokens; vision embeddings sit outside it. On anything self-hosted, the number worth trusting is the one you measure. How far an engine can get with it is no longer hypothetical, though it is still not independent. vLLM implemented the decoder-side trim and Bounded Replay over the three weeks after release and published the before-and-after in the speed section below — the closest thing yet to a check on the specification, and one made by an engine measuring its own work rather than by an evaluator measuring the model.

النتيجة المستقلة، التي أُجريت مرتين: 39 لأوزان سبتمبر، و45 لـ1017

Artificial Analysis has published the first outside evaluation of DeepSeek V4.1 Flash, and it is the number the beta could not produce. On the current version of the Artificial Analysis Intelligence Index (v4.3.2), the model scores 39 in Reasoning mode at maximum effort. On the same scale and the same setting, DeepSeek V4 Pro 0813 scores 36 and DeepSeek V4 Flash 0731 scores 34. Read together, those three rows say the thing DeepSeek's own marketing claimed and its benchmarks could not establish on their own: the smallest model in the family is now the strongest one. Artificial Analysis describes 39 as well above the median of 19 for open-weight models of comparable size. It was, for five weeks, a score for the September checkpoint alone: the evaluator's listing carried the September release date and nothing else. On October 18 that changed, and the change is the news this piece did not have when it first ran. A second row appeared for DeepSeek V4.1 Flash 1017, dated October 17, scored 45 on the same index at the same maximum effort. That puts the 1017 weights level with GLM-5.3 on 45, two points below GPT-5.6 Sol on 47 and six below Claude Opus 5 on 51 — and it makes the checkpoint the highest-scoring DeepSeek model the site has measured, five weeks after the September weights took that title. It also gives the model a standing it did not have before, and one the evaluator states outright in its October 19 write-up: second on the open-weights table, tied with GLM-5.3 and a point behind MiMo-V2.6-Pro at 46, with Step 5 Preview and Kimi K3 a point further back at 44. The six points are not spread evenly across the ten evaluations. The same write-up reports the largest moves in the agentic and knowledge components — a GDPval-AA v2.1 Elo of 1,746 against 1,600, an AA-Briefcase v1.1 Elo of 1,570 against 1,420, and an AA-Omniscience non-hallucination rate of 33% against 4%, with accuracy rising from 43% to 46% and that evaluation's index moving from -5 to +5. Those are the evaluator's own numbers for its own tests, published by it rather than reproduced here, and they describe a model that became markedly less willing to invent an answer — a different improvement from the speed one DeepSeek's changelog leads with.

The comparison that matters here is the internal one, because it is the same comparison DeepSeek's routing table makes. V4.1 Flash carries 552B total parameters with 16B active per token; DeepSeek V4 Pro 0813 carries about 1.6T total with roughly 49B active. A model with roughly a third of the backbone scores three points higher on a composite its maker did not run — nine points higher once the 1017 weights are counted, since the 1.6T flagship has not moved from 36. The size-to-score trade is the entire thesis of the Causal Encoder-Decoder design, and an evaluator with no stake in the result has now produced a number consistent with it.

The price half of the claim holds up too, and this is the first price comparison in the story that does not come from a DeepSeek rate card. Artificial Analysis lists DeepSeek V4.1 Flash at $0.30 per 1M input tokens and $1.20 per 1M output, against $1.32 and $3.96 for DeepSeek V4 Pro 0813 — 4.4× cheaper on input and 3.3× on output, with a blended $0.18 per 1M on the site's 7:2:1 cache-hit, input and output mix. It is also cheaper than the older, smaller DeepSeek V4 Flash 0731 it retires, which the site lists at $0.44 and $1.32 while scoring five points lower. Cheaper than the model it replaces, cheaper than the model it displaces, and ahead of both on the index.

Two caveats belong on the number. It is a composite of ten evaluations run at maximum reasoning effort, so it says nothing about the lower effort settings most production traffic actually uses, and nothing about how quality holds at the far end of a 1M-token window. And it is a single evaluator's score: independent, but not replicated — and the composite itself has been revised since this piece first cited it, from v4.3 to v4.3.2, which moved DeepSeek V4.1 Flash's score from 40 to 39 and its predecessor's from 35 to 34. What it does establish is narrow and useful — the one claim the beta could not support, that a 552B model is genuinely competitive with DeepSeek's own 1.6T flagship, now has a source outside DeepSeek. The 1017 row carries the same two caveats and one more. It is a single evaluator's score, for weights five weeks newer than most of the evidence here, and it is one composite rather than a replication of any vendor row. Its per-task price has also moved on the site: the September checkpoint is listed today at $0.49 per Intelligence Index task, where this piece first carried $0.27, and the 1017 checkpoint at $0.26. Treat the dollar figure as a live number on somebody else's dashboard rather than a constant — which is the same lesson the index score itself taught when it went from 40 to 39. The cost table beside the score is the rest of the caveat, and it cuts both ways. The model one place above it is also half its price: MiMo-V2.6-Pro reaches 46 at $0.13 per Intelligence Index task against $0.26 here, and Artificial Analysis's own comparison puts GPT-6.1 Sol at medium effort on 48 for $0.21. What the 1017 checkpoint does win is the head-to-head DeepSeek has been selling since September — it beats GLM-5.3-Flash on both axes of that table, 45 against 42 and $0.26 against $0.28 a task — which is the pairing where capability and cost move the same way.

النتيجة الخارجية التالية: ARC Prize تبني الاختبار

في 25 سبتمبر، فتح مات مازور، قائد اختبار النماذج في ARC Prize، طلب سحب في مستودع القياس المرجعي ARC-AGI-3 الخاص بالمؤسسة، مضيفًا محول مزوّد DeepSeek من الدرجة الأولى: ملف تعريف واحد، باسم deepseek-v4-1-flash-low-provider-adapter، حقل الطراز فيه هو deepseek-ai/DeepSeek-V4.1-Flash. وصل الادعاء بأن ARC Prize تعتزم تقييم هذا الطراز في منشور واحد على وسائل التواصل، لذا تحققنا منه بالرجوع إلى كود المؤسسة نفسه لا إلى المنشور. المحول وملف التعريف واختباراته الوحدوية ووثائقه موجودة في الفرق (diff)، والمستودع هو إطار الاختبار الذي تسميه سياسة اختبار ARC Prize لمعيارها المرجعي التفاعلي الوكيلي. ما لا وجود له هو نتيجة: طلب السحب مفتوح وغير مدموج، ولا توجد درجة، ولا تاريخ نشر، ولا بيان بشأن مستويات الاستدلال التي ستُنشر. سبب أهميته هو شكل القياس لا خبر أن نتيجة قادمة. يطرح ARC-AGI سؤالاً لا تطرحه اللوحات الأخرى: يمكن أن يكون الطراز سريعًا ورخيصًا وقويًا داخل إطار اختبار، ومع ذلك يحصل على درجة في الستينيات في مهام لم يرها. هذا هو السؤال الذي لم يطرحه أحد على DeepSeek V4.1 Flash بعد.

يحدد الملف التعريفي بدقة كيف سيسير تشغيل كهذا، وستؤثر ثلاثة تفاصيل في الرقم الذي سيُحتسب في النهاية. فهو يشغّل النموذج عند إعداد جهد الاستدلال المنخفض لا الأقصى، ويبث الاستجابات، ويضبط سياق منصة اختبار بحجم 1M توكن مع مُشغّل لضغط الملخص عند 175,000 توكن، ويحدّ سقف الإخراج عند 262,144 توكن. وهو يمر عبر نقطة النهاية المتوافقة مع OpenAI من Baseten بدلًا من نقطة DeepSeek الخاصة، وهو مهيأ بسعر 0.30 دولار لكل مليون توكن إدخال و1.20 دولار لكل مليون توكن إخراج — أي أسعار بطاقة DeepSeek في وقت الذروة، وليس الأرقام خارج وقت الذروة المقتبسة لاحقًا في هذا المقال. وتحسب ARC Prize التكلفة لكل مهمة من الأسعار المضبوطة في الملف التعريفي، لذا سيكون الرقم النهائي محسوبًا بسعر الذروة، وهو أمر يستحق التذكر قبل مقارنته بـ 0.042 دولار لكل مهمة سجّلها DeepSeek V4 Flash 0731 عند اختباره. الميزة غير المعتادة الأخرى هي حل بديل لسلوك وضع التفكير لدى DeepSeek: لا يُعاد إدراج محتوى الاستدلال التاريخي إلا عندما يحمل الطلب أدوات، لذا يجعل المهايئ النموذج يصدر كل إجراء من إجراءات ARC كاستدعاء أداة submit_action، ويعيد إرسال رسالة المساعد نفسها مع إرفاق استدلالها، بدلًا من استخراج الإجراءات من النص السردي.

المعيار الذي سيُقاس عليه هذا يأتي من لوحة ARC Prize الخاصة. إن DeepSeek V4 Flash 0731، وهو نقطة الحفظ التي يُخرجها V4.1 Flash إلى التقاعد، هو المُدخل الموثَّق في اللوحة لفئة Flash التي ينتمي إليها: 61.4% على ARC-AGI-2 عند أقصى جهد استدلالي و89.0% على ARC-AGI-1، بتكلفة 0.042$ و0.021$ لكل مهمة. أما Dots3-Note Preview فيقع عند 76.8% على ARC-AGI-2، وهو الـ77% الذي يشير إليه المنشور الذي نبّه إلى المهايئ، وتوقّع الحساب نفسه — بأن أي نتيجة دون 80% من خلفٍ أكبر بكثير ومتعدد الوسائط أصلاً ستكون مخيبة للآمال — هو تأطيره الخاص لا هدف منشور. ويستحق الجزء المتبقي من ذلك التأطير تصنيفاً: فالادعاء بأن V4.1 Flash يولّد رموزاً أكثر بنسبة 29% من النموذج الذي يحلّ محلّه هو ادعاء غير منسوب وغير محدد، ولا شيء مما نشرته DeepSeek أو ARC Prize يقول ما الذي يقيسه.

One caution about which board this is. The adapter landed in the ARC-AGI-3 repository, the harness for the foundation's interactive agentic benchmark, and the separate repository that runs ARC-AGI-1 and ARC-AGI-2 carries no V4.1 entry at all — its model list has not been updated since July 27 and still names older DeepSeek endpoints. So the datable fact is that ARC Prize has wired this model into its agentic harness; whether it will also appear on the ARC-AGI-2 board, where the 61.4% comparison lives, is not something the public code shows today. The practical reading for anyone choosing a model has widened. Four independent instruments on DeepSeek V4.1 Flash are now in play — a composite index that has been run twice, once per checkpoint; a community kernel board; a harness-controlled cyber benchmark; and, since October 14, a hardware serving benchmark — and two of them put the model inside a harness on tasks it has not seen. The newer of the two is the composite, which has carried Terminal-Bench 4.0 since index v4.3: 66 terminal tasks, pass@1 averaged over three repeats, every model under the same agent harness. Its October 18 result for the 1017 weights is 33.8. The cyber benchmark, below, is the closest any of them comes, and on it this model loses the one head-to-head it makes. The ARC profile is the evaluation built to answer the reasoning question, and it is still the one without a number. Until it runs, the model's reasoning ceiling is unmeasured, and the number to hold it against is its predecessor's.

أول درجة سيبرانية خارجية: إطار الاختبار يحرّك الرقم أكثر من النموذج

في 28 سبتمبر، نشر مختبر التقييم Generality Labs متابعةً لعملها السيبراني السابق بعنوان «هل يهم الهارنس عند تشغيل Cyber Benchmarks؟» — وهي أول قياس خارجي مكتمل يضع DeepSeek V4.1 Flash إلى جانب نموذج آخر مفتوح الأوزان على نفس المعيار السيبراني. المعيار هو Exploit Bench، وهو تقييم من Carnegie Mellon من إعداد Seunghyun Lee والأستاذ David Brumley، ويقيّم إلى أي مدى يتسلق الوكيل سلّمًا من ست عشرة قدرة — من الوصول إلى شيفرة قابلة للاستغلال، إلى إثارة الخلل، إلى بناء بدائيات الاستغلال، إلى تنفيذ شيفرة اعتباطية — في مواجهة V8 الإنتاجي مع تمكين صندوق الحماية الأمني الخاص بـ V8. شغّل المختبر نموذجين، DeepSeek V4.1 Flash وGLM-5.3-Flash من Z.ai، بميزانية 250 مليون توكن لكل مهمة على مدى 10 مهام مسحوبة من أصل 41 مهمة في المعيار بواسطة نظرية استجابة الفقرة، وأعاد تشغيل كل نموذج ضمن سبع تهيئات للهارنس: ست واجهات سطر أوامر للوكلاء — Claude Code وCodex وKimi CLI وReAct وGemini CLI وOpenCode — إضافة إلى الوضع الافتراضي الخاص بالمعيار، الذي يسميه التقرير Original. Original ليس واجهة سطر أوامر تابعة لطرف ثالث على الإطلاق: فلا حدّ للتوكنات فيه، وسقف 300 دورة، ولا ضغط للسياق. الحساب الذي نبّه إلى هذه النتيجة يصف Original بأنه تقريبًا حلقة while في Python؛ وهذا توصيفه هو لا توصيف الباحثين، الذين وصفهم الخاص هو الوارد أعلاه.

النتيجة هي تباين لم تكن أي درجة منفردة لتُظهره. متوسط القدرة من أصل 16. يبلغ DeepSeek V4.1 Flash أفضل حالاته تحت Claude Code عند 9.3، مع Codex في المرتبة الثانية عند 8.9؛ وينخفض النموذج نفسه إلى 7.4 على Kimi CLI، و5.9 على OpenCode، و5.7 على Gemini CLI، و5.3 على ReAct، و4.7 تحت Original — وهو الإعداد الافتراضي للاختبار القياسي الخاص به، والأسوأ بين السبعة. أما GLM-5.3-Flash فيقلب هذا الترتيب عند الطرفين. أفضل إعداد له هو Codex عند 10.6، فوق 10.02 التي تحملها مجموعة بيانات الدراسة نفسها لـ Mythos Preview على مجموعة المهام الكاملة المكوّنة من 41 مهمة، رغم أن الباحثين يحذّرون من أن مجموعة فرعية من 10 عينات ليست ممثلة تمامًا؛ وأسوأ إعداد له هو Claude Code عند 6.2، وهو إطار التشغيل نفسه الذي يمثل أفضل نتيجة لـ DeepSeek V4.1 Flash، بينما يأتي عند 8.9 على Gemini CLI، و8.2 على Kimi CLI، و8.1 على OpenCode، و7.7 على ReAct، و6.4 تحت Original. وبالمتوسط على السبعة جميعًا، يقع GLM-5.3-Flash قرب 8.0 وDeepSeek V4.1 Flash قرب 6.7. لذا فإن كلا شطري الادعاء الذي انتشر مع هذه النتيجة يصمدان أمام البيانات المنشورة: اختيار إطار التشغيل يساوي عدة نقاط، وعلى هذا الاختبار القياسي يُعد DeepSeek V4.1 Flash الأضعف بين النموذجين — باستثناء تحت Claude Code، حيث يتقدم بأكثر من ثلاث نقاط.

يؤكد الباحثون صراحةً أنهم لا يستطيعون تفسير هذا الانتشار تفسيرًا كاملًا. وخيطهم الإيجابي الوحيد هو الطريقة التي يضغط بها كل هارنس السياق، والفارق مسألة صياغة. فمطالبة الضغط في ReAct تطلب من النموذج أن «يكتب ملخص استكمال يتيح لك (أو لنسخة أخرى من نفسك) استئناف العمل»؛ وفي النصوص التي نشرها المختبر، يؤدي ذلك أحيانًا إلى أن يعلن النموذج انتهاء العمل، وبعدها يوقظه الهارنس مجددًا فيعيد إرسال التقرير نفسه عشرات المرات حتى تنفد الميزانية. أما Codex فيصوغ اللحظة نفسها بوصفها «أنشئ ملخص تسليم لنموذج لغوي كبير آخر سيستأنف المهمة»، ويفترض المختبر أن مخاطبة وكيل مختلف هي ما يُبقي النموذج يُحرز تقدمًا على مدى ميزانيات طويلة. وهم يسمّون هذا فرضيةً لا سببًا. وثمة خطأ في الإعداد يهم أيضًا عند قراءة الجدول: فقد ضُبط الضغط في ReAct على 100 ألف توكن بدلًا من عتبة التشغيل المقصودة، فكان يضغط أكثر بكثير مما اقتضاه التصميم، وهذا ما يخفض صفَّي ReAct لكلا النموذجين. وتجدر مرافقة أي رقم مأخوذ من هذه الدراسة بتحذيراتها هي نفسها — نموذجان، و10 عينات لكل إعداد هارنس، و70 عينة سؤال لكل نموذج مقابل 41 مهمة أصلية في المعيار المرجعي، وثلاث عينات أخطأت أو لم تنتهِ قط، اثنتان منها GLM-5.3-Flash. ويذكر المؤلفون أن العينة الصغيرة مشوبة بالضجيج بطبيعتها، وأنه لا ينبغي افتراض أن النتيجة تمتد إلى مجالات ومهام أخرى، وأن عملًا سابقًا لـ METR لم يجد فروقًا دالة إحصائيًا بين الهارنسات في المهام طويلة الأفق. إنه مختبر خارجي يشغّل معيارًا مرجعيًا ليس ملكه، وليس أصحاب المعيار أنفسهم، وهو غير محكَّم. لكنه ليس رقمًا من مورّد، وهو أول رقم خارجي في هذا المقال يناقض رواية المورّد نفسها بدل أن يكتفي بالإخفاق في تأكيدها.

ثمة أمر واحد هنا يمكن للموجّه أن يتصرف بناءً عليه، وآخر لا يمكنه ذلك. لا يمكنه اختيار هيكل التشغيل (harness) الخاص بك — فذلك هو حلقة الوكيل التي تشغّلها بالفعل، وبيت القصيد في هذه الدراسة أن هذا الاختيار جوهري وحاسم. ما يمكنه فعله هو جعل نصف النموذج في الثنائي تغييرًا بسطر واحد. DeepSeek V4.1 Flash وGLM-5.3-Flash كلاهما قابل للتوجيه عبر OrcaRouter بمفتاح واحد متوافق مع OpenAI، وبسعر القائمة الخاص بالمزوّد نفسه مع هامش ربح مضاف نسبته 0%، لذا يمكن لفريق يريد إيجاد مزيجه الخاص من النموذج وهيكل التشغيل أن يبدّل النموذج تحت هيكل تشغيل ثابت دون عقد ثانٍ أو تكامل ثانٍ، وأن يترك خلفه نموذجًا مُثبتًا كخيار تجاوز فشل تلقائي إذا تبيّن أن التبديل تراجع. وهذه هي الحجة نفسها التي يقدّمها قسم التبني أدناه، مطبَّقة على قرار جعلته دراسة هيكل التشغيل ملموسًا للتو: استنادًا إلى هذه الأدلة، قد يكون هيكل التشغيل الذي تشغّله أكثر أهمية من النموذج الذي تختاره، وجانب النموذج في ذلك الثنائي هو الجانب الذي يمكنك تغييره في فترة بعد ظهر واحد.

أين يقع مقابل الحدود المتقدمة، وأين لا يزال يخسر

The September model card compares V4.1 Flash against Opus-5.0, GPT-5.6 Sol, Kimi K3 and GLM-5.3 at maximum reasoning effort, with its own DeepSeek V4 Pro and DeepSeek V4 Flash as internal baselines. Every figure below is vendor-reported, run on DeepSeek's harnesses, and unreproduced by any third party. The card notes that agentic results use the Minimal mode of DeepSeek Harness with a 1M-token context for code benchmarks and a 512K window for visual ones, so these are not directly comparable to published third-party numbers on the same benchmarks. The independent scores in the sections above are different measurements on different sets of evaluations; they support the general picture and corroborate no individual row below, and the October 17 rows in the section above come from a different card and a different run of the same benchmarks.

في الأعمال الوكيلية والبرمجية، تضع أرقام المورّد V4.1 Flash فوق النماذج المتقدمة التي يسميها:

• Terminal-Bench 2.1 — 90.6، مقابل 89.1 لـ Opus-5.0، و88.8 لـ GPT-5.6 Sol و87.9 لـ DeepSeek V4 Pro

• DeepSWE v1.1 — 74.2 محلولة، مقابل 74.0 و73.0 و62.7

• CyberGym — 88.1، مقابل 84.5 لكل من GPT-5.6 Sol وGLM-5.3، و83.3 لـ DeepSeek V4 Pro

• AutomationBench — 54.8، مقابل 50.3 لـ Opus-5.0 و37.7 لـ DeepSeek V4 Flash

• Agent's Last Exam — 31.8، مقابل 28.6 و25.7

• Codeforces — تقييم 3,471، متقدّمًا على DeepSeek V4 Pro بتقييم 3,348 وDeepSeek V4 Flash بتقييم 3,289

• MathArena Apex — 65.6، بمستوى Kimi K3 ومتقدّم على DeepSeek V4 Pro عند 65.3

وعلى المجموعات الوكيلية الأكثر صعوبة، فإنه لا يفعل ذلك:

• Terminal-Bench 3.0 — 30.0، متأخر كثيرًا عن Opus-5.0 عند 43.3 وGPT-5.6 Sol عند 34.4، لكنه متقدم بفارق كبير على DeepSeek V4 Pro عند 11.8

• Terminal-Bench 4.0 — 31.2، مقابل 51.8 و39.9، ومتفوق مجددًا على DeepSeek V4 Pro بفارق 12.4

• ProgramBench — 20.3، مقابل 37.0 لـ Opus-5.0

• NL2Repo-Bench — 64.0، مقابل 75.3 لـ Opus-5.0

• HLE — 36.8، مقابل 56.3 و44.5

• ExploitGym — 15.3، مقابل 33.7 و22.1

The shape of that list is more informative than any single row. V4.1 Flash clears the frontier on benchmarks that reward sustained tool use inside a harness, and loses on the ones that reward depth of reasoning — the pattern you would expect from a model engineered around throughput and context economics rather than around raw reasoning ceiling. The internal comparison is the striking one: it beats DeepSeek V4 Pro on essentially every row, including the rows where both lose to Opus-5.0. DeepSeek is not claiming its cheapest model is the best model on the market. It is claiming its cheapest model is the best model DeepSeek makes, which is a claim its own routing table backs up.

كم تبلغ تكلفته، وكيف تبدو المقارنة مع Pro في الواقع

September 10's rate card is now simply the V4.1 Flash card, published on DeepSeek's Models & Pricing page with a peak and off-peak split:

• الإدخال، إصابة ذاكرة التخزين المؤقت — 0.003 دولار لكل مليون رمز في غير أوقات الذروة، و0.006 دولار في وقت الذروة

• الإدخال، فقدان ذاكرة التخزين المؤقت — $0.15 لكل مليون رمز خارج أوقات الذروة، $0.30 في أوقات الذروة

• الإخراج — ‏0.60 دولار لكل مليون رمز خارج أوقات الذروة، و1.20 دولار في أوقات الذروة

• نوافذ الذروة — 01:00–04:00 و06:00–10:00 بالتوقيت العالمي الموحد، من الاثنين إلى الجمعة؛ كل ساعة أخرى تُفوتر بنصف سعر الذروة

• بالرنمينبي — ¥0.02، و¥1 و¥4 خارج أوقات الذروة، لتتضاعف إلى ¥0.04، و¥2 و¥8 في أوقات الذروة

Compare that with the model it is displacing. DeepSeek V4 Pro bills $0.022 per 1M cache-hit input, $0.66 cache-miss input and $1.98 output off-peak — roughly 4.4× on cache-miss input and 3.3× on output. Those are the numbers the September 14 rerouting would have applied to every deepseek-v4-pro call, which is why the plan was a price story as much as a capability story, and they are still the reason the comparison matters: DeepSeek withdrew the reroute, not the rate card. Artificial Analysis reaches the same arithmetic from its own listings — $0.30 and $1.20 per 1M for V4.1 Flash against $1.32 and $3.96 for DeepSeek V4 Pro 0813, the same 4.4× and 3.3×, and a blended $0.18 per 1M on the site's 7:2:1 cache-hit, input and output mix. The card also undercuts DeepSeek V4 Flash 0731, the model it retires, which the same site lists at $0.44 and $1.32.

سعر الرمز الواحد ليس تكلفة المهمة الواحدة، والنسخة التجريبية هي الدليل على ذلك. أبلغ مختبِرون في اليوم الأول أن جلسات مدتها خمس دقائق تستنزف رصيدهم على النموذج الجديد أسرع من المدة الزمنية نفسها على DeepSeek V4 Flash، لأن النموذج مصمم لإصدار الرموز بسرعة، وتُحسب الفاتورة بالرموز لا بالدقائق. بطاقة يبلغ سعرها ربع سعر Pro تكون صفقة رابحة إذا أُنجزت مهمة بعدد مماثل من الرموز؛ وتكون بلا فرق إذا كان النموذج يفكر مدة أطول للوصول إلى ذلك.

Where an outside number does exist on that question, it points both ways. Artificial Analysis puts DeepSeek V4.1 Flash at $0.27 per Intelligence Index task on its own evaluation set — a real per-task figure, produced outside DeepSeek, and a low one. But the same run generated 250M output tokens against a median of 140M for comparable models, and the site's summary calls the model very verbose in the same breath as calling it fast. Verbosity is free inside a benchmark harness and expensive in production, which is precisely the mechanism the beta testers were describing when they reported balances draining faster in wall-clock terms. The honest state of the per-task question is that one independent evaluator has now measured both halves of it, on its own evaluation set, and the answer still depends on what you are running. The composite cannot show where the extra output goes, and the beta testers never said either. The first hands-on reports do, and they are in the speed section below: the model is verbose because it is checking. The October 17 update is the first time DeepSeek has claimed to move that number: the changelog says the new checkpoint “reduces task runtime and output length,” and the post a few hours later says the same thing in fewer words — “faster task completion, and shorter outputs.” That is DeepSeek addressing, on its own authority, the exact quantity the outside measurement complained about, with no figure attached to either statement. The first outside look at it, on October 18, found the aggregate volume down — 210 million output tokens across the index against 250 million for September — and the per-task figure flat, at 89,495 tokens against 88,574.

تكلفة الشهر: بدل بقيمة 80 دولارًا، ثم طبقة مجانية

تجيب البطاقة أعلاه عن تكلفة التوكن. وثمة رقم مختلف تغيّر مرتين، وفي المرتين جاء من OpenCode — وكيل البرمجة الذي يسعّر اشتراكه Go الوصولَ إلى هذا النموذج كبدل استخدام شهري، لا كتوكنات محسوبة بالعدّاد — متحدثًا عن اشتراكه الخاص لا عن النموذج. في 25 سبتمبر نشر حساب المنصة: «عملية Cheepseek المرحلة 2: أصبح استخدام بقيمة 60 دولارًا لـ DeepSeek v4.1 Flash دائمًا الآن.» وفي 12 أكتوبر نشر التكملة بالصيغة نفسها — «عملية Cheepseek المرحلة 3: استخدام بقيمة 80 دولارًا لـ DeepSeek V4.1 Flash» — وكانت صفحة التسعير الخاصة به تقرأ 80 دولارًا لهذا النموذج حيث كانت تقرأ 60 دولارًا في أواخر سبتمبر. وأُخذت تلك القراءة مرة ثانية في 13 أكتوبر، من الصفحة نفسها، فأعادت الصف نفسه: 80 دولارًا على فئة Go البالغة 10 دولارات شهريًا، و160 دولارًا على Go Plus. لذا فليس هذا الرقم ادعاء منشور واحد عن نفسه؛ بل هو سعر منشور من منصة واحدة لحمل هذا النموذج، وكان لا يزال موجودًا بعد يوم. وعند قراءة المنشورين معًا، يقولان ما انتهى إليه معنى «دائم»: لا تاريخ انتهاء — ورقم أعلى بعد ثلاثة أسابيع. وما ليس عليه هذا التغيير هو كونه طريقة جديدة للوصول إلى النموذج — فهي تأتي بعد ثلاثة أيام، في الفقرة أدناه. لقد كان DeepSeek V4.1 Flash يستجيب عبر واجهة API الخاصة بـ DeepSeek منذ 10 سبتمبر، ويحمله OrcaRouter بأسعار DeepSeek نفسها منذ الإصدار، لذا فإن الأوزان والبنية وكل صف من صفوف القياس في هذا المقال لم تمسّها إعادة تسعير الاشتراك؛ ما غيّره أكتوبر هو مَن يرغب في حمل النموذج وبأي شروط.

المُدخل المجاني تغيير من نوع مختلف عن أيٍّ من هذين، وهو الوحيد من بين الثلاثة الذي يغيّر ما يمكن للقارئ فعله هذا الأسبوع. في 15 أكتوبر، أضاف OpenCode مُدخلًا منفصلًا لهذا النموذج — DeepSeek V4.1 Flash Free، بمعرّف النموذج deepseek-v4.1-flash-free — يُقدَّم دون مقابل. تُظهره قائمة النماذج العامة للمنصة بجانب المعرّف المفوتر، وتحمل وثائق Console الخاصة بها بندًا للنماذج المجانية يصفه بأنه مجاني لفترة محدودة، ويصف مزوّده بأنه يتبع سياسة عدم الاحتفاظ (zero-retention) لا تدرّب على البيانات، ويضع جدول الأسعار المنشور الخاص بها الإدخال والإخراج والقراءات المخزنة مؤقتًا عند Free. لكنه ليس المنتج الكامل. الحدود المفهرسة للمُدخل المجاني — التي أُضيفت إلى models.dev، الكتالوج الذي تُستمد منه قائمة نماذج OpenCode نفسها، في 15 أكتوبر — تسمّي المُدخل المفوتر بوصفه النموذج الأساسي، وتضع سقف النافذة عند 200,000 توكن مع سقف إخراج يبلغ 32,000 توكن، مقابل سياق 1M وحد أقصى للإخراج 384K يحمله مُدخل deepseek-v4.1-flash المفوتر. وهذا هو نفس الأوزان خلف شروط تقديم أضيق، لا نموذج أصغر أو مُقطَّر، وسقف الإخراج هو البند الذي يؤثر أولًا في العمل الوكيلي، حيث يمكن أن يتجاوز تعديل ملف واحد هذا السقف. ولا ترفق المنصة أي تاريخ انتهاء بهذا الترتيب، ولا تنشر بطاقة أسعار له، لأنه لا توجد تسعيرة. لا شيء من هذا جاء من DeepSeek: إنه قرار تجاري لمنصة واحدة بشأن نموذج لا تملكه، مذكور في قائمة نماذجها ووثائقها وجدول أسعارها، وهو المصدر الوحيد لكل ذلك — ولم نجد أي قياس مستقل لكيفية تصرف نقطة النهاية المجانية تحت الحمل، و«فترة محدودة» عبارة المنصة لا جدول زمني. اقرأه على أنه أرخص طريقة لوضع نافذة من 200,000 توكن أمام هذا النموذج اليوم، وليس كتغيير في ماهية النموذج.

The mechanics come from that pricing page, read on October 12 and re-read on October 13. Go is a $10-per-month subscription, and its table pairs every model on it with an estimated request count per five hours and a monthly usage limit, in separate columns for the two tiers. DeepSeek V4.1 Flash carries the largest allowance on the page: $80 a month and 34,670 estimated requests per five hours on Go, doubling to $160 and 69,330 on Go Plus. The rows bill the model at DeepSeek's own card — $0.15 per 1M input tokens, $0.60 output and $0.003 for a cached read at off-peak, doubling to $0.30, $1.20 and $0.006 inside the weekday peak windows — and those are the vendor's figures, unchanged since the September reading this section already carried. Two things the page showed then are not on it today, and both are worth flagging because the earlier reading relied on them: the three-bucket draw-down rule — 20% over any five hours, 50% across a week and 100% across a month — no longer appears, and the documentation URL the request estimates were originally read from, opencode.ai/docs/go, returned 404 when we checked. The request counts above are the platform's arithmetic on a call shape it does not publish, not ours.

The change is checkable against the page, and so is the fact that there has now been more than one. In late September the same rows read $60 with no expiry attached, after a promotional $15 allowance was multiplied four times to $60 and carried September 20 as the date the multiplier would lapse; the request estimates were published in two columns, 6,500 / 16,250 / 32,500 standard against 26,000 / 65,000 / 130,000 with the multiplier applied. The rows read $80 on both readings, and the published estimates for this model are 34,670 per five hours and 69,330 a month — a higher dollar allowance sitting beside a lower monthly request count than September's expanded set, which cannot both describe the same assumed call shape. The page states no token mix and offers no explanation, so the difference is a fact about the two readings rather than something we can resolve from outside. The rates themselves were never a discount on DeepSeek's card — per-token pricing has been the vendor's throughout, and it still is. What an allowance does not do is say anything about quality: it is one platform's judgment that it can carry this model's traffic at those rates without a deadline, which is evidence about the serving ecosystem rather than about the model.

The distinction matters when you are choosing, because a flat ceiling and a pass-through meter are different products. DeepSeek V4.1 Flash is on OrcaRouter at DeepSeek's own rate with 0% markup — $0.15 per 1M input and $0.60 per 1M output off-peak, the same numbers the card above lists, with nothing bundled around them: no subscription to buy and no monthly allowance to exhaust. An $80 ceiling on a fixed monthly price is the right shape for one developer running an agent all day, where the question is what the month costs — and a ceiling already raised once, by a third, is also a reminder that the figure is a commercial decision rather than a property of the model. A free tier with a 200,000-token window is the right shape for trying the model this week and the wrong shape for the codebases that need the full window, which is the one thing the two tiers of the same model do not share. A vendor rate passed through unchanged is the right shape for a service, where the question is what each call costs and whether the provider's next price move reaches you the day it happens. Both answer what this model costs; they answer it for different spenders.

السرعة: أرقام النسخة التجريبية، وادعاء الصانع، وقياس المحرّك نفسه قبل وبعد.

عندما فُتحت النسخة التجريبية، كانت الإنتاجية هي العنوان الأبرز. أفاد المطورون بنحو 420 رمزًا مُخرَجًا في الثانية في التوليدات الطويلة، مع ذُرى تتجاوز 500، وأنتجت عملية تشغيل جرى تشاركها على نطاق واسع أكثر من 71,000 رمز في أقل من ثلاث دقائق. تظل تلك قياسات مجتمعية أُجريت في ظروف غير خاضعة للتحكم، وينبغي أن تُقرأ على هذا النحو — لكن الآلية الكامنة وراءها أصبحت الآن موثّقة بدلًا من أن تكون مجرد تخمين. إن ذاكرة KV المؤقتة بحجم 890 بايت لكل رمز ومسار فك ترميز يفعّل 16 مليار معامل هما تصميم مبني من أجل الإنتاجية، لذا فإن البنية على الأقل في الشكل المناسب لإنتاج أرقام كهذه. أما حجمها الحقيقي فهو مشكلة الفقرة التالية، والجواب أصغر.

DeepSeek's own claim, in the changelog, is that V4.1 Flash “has comprehensively surpassed V4 Pro in performance, cost, speed, and total time.” The word doing the work there is total time: it is the vendor framing speed as task completion rather than tokens per second, which is the metric that survives contact with a real agent loop that writes a file, reads the error and tries again. Artificial Analysis now puts a measured figure on the older claim: 185 output tokens per second for DeepSeek V4.1 Flash, against 175 for DeepSeek V4 Flash 0731. That is a real third-party number and it is roughly half the 420 tokens per second developers reported during the beta, which is why the beta figures should stay where they belong — in the uncontrolled-conditions column.

On October 7 the vLLM project published its accounting of the three weeks since release, and it is the first measured speed figure in this piece that is neither a token counter nor a vendor slogan. Writing as Inferact and the vLLM Team, the project reports that DeepSeek V4.1 Flash on its current build runs 1.9× faster than its own day-0 implementation at low concurrency and delivers 5.3× the throughput under a 150 TPS constraint — the figure its summary carries, against “about 5×” in the body of the same post. Both are measured on the SemiAnalysis AgentX agentic serving benchmark, with TP4 and FlashInfer attention behind the low-latency number and DEP2 with data-parallel attention behind the throughput one, on NVIDIA GB200 and GB300 hardware. Three things about that result are worth keeping straight. It compares vLLM against vLLM: the baseline is the project's own day-0 recipe, not another engine and not another model, so the number says how much the serving stack improved rather than how this model compares with anything else. It is the runtime's own account of its own work, and no third party has reproduced it. And the largest single gain comes from the runtime rather than the weights — SWA bounded replay, which reruns only the last 128 tokens of a sliding window instead of replaying that window across layers, now ships on by default for this model behind a single flag. The project reports roughly 30% off time to first token from replay plus CUDA-graph capture, a 30–40% cut in prefill computation from the decoder-side trim, and nearly 70% off TTFT at around 100K throughput once three optimisations stack. It also reports where replay hurts: without CUDA graphs it is slower than the baseline on short prompts, up to 12% worse at 1K tokens, because launch overhead outweighs the work saved. Accuracy was checked on GSM8K and GPQA with no meaningful difference reported, gaps within about 1.5 standard errors — the figure to hold against the report's own statement that bounded replay is approximate rather than exact. The project adds that the model needed no KV-cache offloading anywhere in the benchmark and expects offloading to start paying only at higher concurrency with prefill and decode split across machines. This is a different measurement from the kernel board below, which scores kernels the model writes; vLLM's figures are about what the model costs to serve once an engine has implemented its architecture, and the engine here is the one reporting. The project posted again on October 17, the day DeepSeek swapped the checkpoint, and the number it led with was a different kind of figure: 1.3B tokens per dollar of total cost of ownership on SemiAnalysis AgentX, at 830,000 tokens per second per GPU counting cached, input and output tokens together, while holding 100 output tokens per second for every user. That is a cost-of-serving result rather than a speed delta, and the serving-cost section below reads it in full. Three things about it bear on this paragraph: it is the same project, the same model and the same benchmark three weeks on from the October 7 accounting; the project frames that accounting as chapter one, at 5x the agentic throughput of day zero, with a next chapter still to come, and no second DeepSeek post has appeared on its blog since; and a cost curve is still the runtime's side of the ledger, not an outside verdict on the model. That distinction is also why this number matters to someone who will never run a node. None of the work is work a caller does: a serving stack that gets 1.9× faster at low concurrency makes the requests it handles faster, and reaching this model through an API is how most teams will benefit from that without owning the hardware. DeepSeek V4.1 Flash is on OrcaRouter at DeepSeek's own rate with 0% markup, on the same OpenAI-compatible key as the rest of the line, with a proven model behind it as automatic failover — the model and the pricing move at the provider's pace, and the choice of engine stays on the provider's side of the line.

ما تكلفة الخدمة: أول أرقام مستقلة للأجهزة

كل ما سبق يقيس النموذج. في 14 أكتوبر، أضافت Artificial Analysis نموذج DeepSeek V4.1 Flash إلى AA-AgentPerf، وهو المعيار الذي تشغّله مقابل أنظمة المسرّعات لا مقابل النماذج، والنتيجة هي أول إجابة علنية على سؤال لا تستطيع قائمة الأسعار حسمه: كم من هذا النموذج يحمله ميغاواط واحد من طاقة مراكز البيانات. يعيد AA-AgentPerf تشغيل جلسات وكلاء برمجة مسجّلة — استدلال متعدد الأدوار، واستدعاءات أدوات، وسياقات تنمو لتتجاوز 100 ألف رمز، مع احتساب إصابات ذاكرة التخزين المؤقت للمطالبات — مقابل حزمة خدمة حيّة، ويحدد أكبر عدد من الوكلاء المتزامنين يستوعبه نظام ما بينما يظل كل واحد منهم ملتزمًا بهدف سرعة لكل وكيل. بالنسبة لهذا النموذج، العنوان الأبرز هو مكوّن NVIDIA بحجم الرف. عند 100 رمز إخراج في الثانية لكل مستخدم، يتصدر GB300 NVL72 المجال بنحو 261 مليون رمز في الثانية لكل ميغاواط، متقدمًا على B300 بنحو 220، وB200 بـ 198 مليونًا، وGB200 بـ 178 مليونًا، وMI355X من AMD بـ 128 مليونًا. هذه أرقام Artificial Analysis، مأخوذة من صفحة الاستدلال في مراكز البيانات الخاصة بها اليوم، وهي قياساتها لأجهزة وتكوينات خدمة محددة وليست قياسات مورّد. وهناك تفصيلان يستحقان الذكر. تُحسب الأرقام لكل ميغاواط من طاقة المسرّع المقيسة أثناء التشغيل — شريحة GPU زائد HBM — وليس من قدرة التصميم الحرارية (TDP) الاسمية. وكل نقطة على المخطط تسمّي المحرّك والإصدار والصيغة الرقمية الكامنة وراءها، لذا فإن الرقم ادعاء يتعلق بحزمة خدمة وليس بالأوزان فقط.

The reason this model's chart looks unlike the others is its size. DeepSeek V4.1 Flash fits on as few as two accelerators, which is why a rack-scale system gets less of an advantage here than it does on a larger model: Artificial Analysis reports B300 staying within about 20% of GB300 NVL72 across the curve and leading it at both ends — the most throughput per megawatt when per-user speed is allowed to fall, and the fastest single agent when it is not. At batch size one, every Blackwell part and the MI355X land between roughly 600 and 670 tokens per second for a single user, against 372 to 392 on H100 and H200. Set against DeepSeek V4 Pro 0813 on the same axis, the smaller model returns about 1.3× the throughput per megawatt on GB300 and about 6.2× on B200 at the 100 tok/s mark — the efficiency claim the architecture section makes from the KV-cache side, arriving from a third party that measured a system rather than a specification. The configurations behind those numbers show the design's central trick in production rather than only on paper: on both the B300 and the MI355X two-accelerator runs, the 196B Engram table is served out of pinned host memory through vLLM's CPU-offload path instead of sitting on the accelerator — the mechanism the release notes described, now visible in the setup behind a published measurement. A second cost figure, in a different currency, arrives from the benchmark the October 7 vLLM post was measured on. On October 17 the vLLM project published a chart putting DeepSeek V4.1 Flash at 1.3B tokens per dollar of total cost of ownership on SemiAnalysis AgentX, with separate curves for GB300 NVL72 under Dynamo, GB200 NVL72, B200, H100 and MI300X. The cost tier behind it is labelled owning at large-hyperscaler volume, and the chip-hour assumptions underneath run from $0.95 for an MI300X and $1.17 for an H100 to $1.73 for a B200 and $2.31 for a GB300, drawn from SemiAnalysis's own July 2026 pricing survey. Two limits belong with it. It is the vLLM project reporting SemiAnalysis's cost model, and neither party has had it replicated. And the chart is dated October 16, the day before DeepSeek replaced the weights behind the model name, so it too describes the September checkpoint. That makes three currencies in this piece for one model: tokens per second per user, tokens per second per megawatt, and tokens per dollar. The last is the one a buyer without a datacenter can act on, because it is the quantity a provider's cost improvement eventually reaches them as. It remains a serving measurement and not a quality one: nothing here says how good the model is, and one leaderboard's hardware results are not a replication of any benchmark row above. What it does establish is that the cost side of this model's rate card now has outside evidence under it. A model a provider can serve at several times the throughput per megawatt is a model a provider can afford to keep cheap, and DeepSeek V4.1 Flash is on OrcaRouter at DeepSeek's own list price with 0% markup, which is the shape that passes a provider's cost improvement through instead of holding it.

كم يكلّف المحرك على رقائق مورّدين اثنين

Until this week the throughput question about this model was answered only by other people's measurements — a runtime accounting for its own work, a benchmark site measuring a system, developers counting tokens in a beta. DeepSeek opened DeepInfer, the framework it says powers this model's production service, under Apache-2.0 this week — the repository appeared on October 16 and the first commit landed on October 20 — and shipped with it a decode Pareto plot for the 1017 checkpoint measured on 64 SM100 accelerators inside one scale-up domain. The configuration printed on the plot is the part that matters, because it is the configuration the comparison below turns on: expert parallelism of 64 over data parallelism of 64, the “384+64 experts” the plot prints, a 131,072-token context, DSpark draft length fixed at three tokens, and a per-token acceptance rate of 75%. The curve tops out just under 30,000 output tokens per second per accelerator at its least interactive end, falls through the mid-20,000s as per-user speed rises past 100 tokens per second, and is down to a few thousand by the chart's right edge. It is DeepSeek measuring its own engine on its own model, so it is not an independent number — but it is the first decode curve for this architecture published by the lab that built it, and the sweep script that produced it is in the same repository.

The write-up published beside the code supplies the operating point, and it is the reason the numbers below are quoted at 230 output tokens per second rather than at some abstract maximum. DeepSeek's own production figures put decode at 14K tokens per second per accelerator while the service holds 230 output tokens per second per user — 300 in coding workloads — against a 96% input cache hit rate, an average input of 77K tokens, a 400 ms mean and one-second p99 time to first token, and prefill provisioned at close to a 1:1 ratio with decode, across a single cluster it says serves 155 trillion tokens a day. The revenue line beside those figures, $38 per accelerator-hour at off-peak pricing, is the first cost number DeepSeek has attached to serving this model publicly. None of it is audited, all of it is the vendor describing its own service, and it is still the most concrete picture of what this architecture costs to run that exists.

The other half of the comparison comes from Huawei's silicon, and it is the weaker-sourced half. A benchmark sheet dated September 30 and headed “Ascend 950DT DeepSeek-V4.1-Flash Benchmark” reports the model deployed across 32 chips at FP8–FP4 mixed precision, with data-parallel attention and expert-parallel mixture-of-experts, on a 128K-token sequence, using DSpark speculative decoding with a forced expert-parallel load-balancer configuration and an average accepted length of 4.1 tokens. It carries three points. Its latency column is time per output token, so the reciprocal of each entry is that row's per-user speed, and that is what lets the sheet be laid against the curve:

• دفعة عالمية 1,536، و9.41 مللي ثانية لكل رمز إخراج — نحو 106 رموز إخراج في الثانية لكل مستخدم — و5,102 رمز في الثانية لكل شريحة.

• دفعة عالمية بحجم 384، و4.40 مللي ثانية لكل رمز إخراج — نحو 227 رمز إخراج في الثانية لكل مستخدم — و2,727 رمزًا في الثانية لكل شريحة.

• دفعة عالمية بحجم 2,560، و13.79 مللي ثانية لكل رمز إخراج — نحو 73 رمز إخراج في الثانية لكل مستخدم — و5,800 رمز في الثانية لكل شريحة، وهو أعلى رقم في الجدول والرقم الذي يستخدمه عنوانه الرئيسي.

Laid against the NVIDIA curve, the middle two rows are the ones that bite, because they sit at the speeds a service actually runs at. The thread that surfaced the sheet reads the comparison as 25,500 tokens per second per accelerator against 5,102 at 100 output tokens per second per user, and 20,000 against 2,727 at 230 — a gap of five times at the first operating point and 7.4× at the second. Reading the same published curve directly moves those NVIDIA figures around: a pixel pass over the image lands nearer 28,000 at the 100-token mark and nearer 17,000 at 230, on either side of the thread's numbers; the band they give is about five and a half times at the first operating point and about six at the second, which narrows the thread's range without changing its shape. That is what a benchmark read off a chart is worth, and it does not change the direction of the result. It is also worth noting that the sheet's 5,800 headline is its least interactive row — the point where NVIDIA's curve is closest to its own peak, and where the ratio is back to about five.

I could not find that sheet on DeepSeek's or Huawei's own pages; it reached this piece as an image attached to a thread on X, and the three Ascend figures should be read as an unpublished vendor benchmark until somebody posts the run behind them. What is checkable is the asymmetry in the configurations, and that comes from DeepSeek's own documentation. DeepInfer's deployment notes recommend prefill at expert parallelism of 2 to 8 and decode at 64 to 128, on “ordinary 8-GPU machines for prefill, ordinary 64–128-GPU supernodes for decode.” On NVIDIA that is a single NVLink domain, which is what the plot above was measured inside. On Ascend, DeepSeek's own communication library says plainly what is not finished: “Larger EP sizes and combine remain under optimization,” with combine carrying extra local reduction overhead and HBM contention against the URMA path, and Huawei firmware work still to come — a caveat public in DeepEP-Ascend since its first release on September 29. That is the fair way to read the gap: the NVIDIA number comes from the shape DeepInfer was tuned for, and the Ascend number from a stack whose own author says the wide-EP path is still being worked. Some of five to seven times is the chip, some of it is the interconnect and the software under it, and nobody has separated the two.

What the gap is worth in money depends on prices nobody has published, and the thread runs the arithmetic anyway: at roughly $139,000 per Blackwell accelerator, extrapolated from grey-market rack pricing, against a rumored $37,000 for a 950DT, NVIDIA has to deliver about 3.8× the throughput per chip to match Ascend on acquisition cost alone; the thread puts the crossover near 6,800 tokens per second per chip. The published Ascend points sit below that line; the thread's forecast that the wide-EP path lifts them to 8,000–10,000 sits above it, and that is a forecast rather than a measurement. Every input is soft — a grey-market price, a rumored one, and figures read off a curve — and none of it accounts for networking, power, cooling, maintenance, or the value of being able to buy the hardware at all.

For anyone who will never rack a node, the useful part is not the scoreboard. It is that the cheapest model in DeepSeek's line is now documented as cheap to serve on two vendors' silicon, with the vendor's own engine published and the numbers behind it downloadable — which is the kind of evidence a price rests on. It changes no rate card: DeepSeek V4.1 Flash is on OrcaRouter at the provider's own list price, $0.15 per 1M input tokens and $0.60 per 1M output at off-peak rates, passed through with 0% markup, so when a provider's serving cost falls the saving arrives as a price that does not have to move rather than as a resale margin. A serving cost argued over on two vendors' hardware is a price with more room under it than one nobody has measured at all.

هندسة النواة: أول لوحة نتائج ليست لـ DeepSeek

صفوف المعيار أعلاه هي أرقام أدوات الاختبار الخاصة بـ DeepSeek نفسها، أما النتائج الخارجية الموجودة الآن فتقيس الاستدلال العام والاستغلال السيبراني، لا البرمجة منخفضة المستوى. كما توجد أول نظرة من طرف ثالث إلى هندسة النواة لدى النموذج، وقد وصلت في شكل لم تكلّف DeepSeek بإعداده: معيار وكيلي يديره المجتمع يطلب من النماذج كتابة نُوى CUDA حقيقية ويقيّم النتائج مقابل roofline — الحد الأقصى النظري للعتاد. لوحة الصدارة هي kernelbench.com، وتعمل على RTX PRO 6000، وهي مستقلة عن كلا الموردين. منهجية التدقيق الخاصة بها مُبلَّغ عنها من المجتمع وليست خاضعة لمراجعة الأقران، لذا اقرأ نسبها المئوية كإشارة اتجاهية من طرف ثالث، لا كقياس معتمد؛ وهي أيضًا أول لوحة نتائج في هذه القصة تجلس فيها أرقام DeepSeek إلى جانب أرقام منافس مباشر.

الصف الرئيسي هو الصف الذي بُنيت حوله البنية المعمارية. الانتباه المتناثر الخاص بـ DeepSeek V4.1 Flash — مكدّس CSA2 الذي ينتج ذاكرة تخزين مؤقت بسعة 890 بايت لكل رمز — كُتب كنواة CUDA بواسطة النموذج، وبلغ 50.19% من roofline المكافئ للنمط الكثيف، محتلًا المركز الرابع على اللوحة خلف Claude Fable 5.1 بنسبة 106.27%، وClaude Opus 5 بنسبة 103.67%، وClaude Fable 5 بنسبة 72.66%. أول ما يُلاحظ أن البنية المعمارية الخاصة بالصانع هي النواة التي يتفوق فيها: فالنموذج المدرَّب على تصميم انتباه متناثر يكتب نواة الانتباه المتناثر بكفاءة أعلى من أي شيء آخر على اللوحة دون المراكز الثلاثة الأولى. والثاني هو المقارنة التي أشار إليها المُخبر. محاولة GLM-5.3 Flash للنواة نفسها تبلغ 4.45% من roofline — أي أقل بمرتبة قدرية من V4.1 Flash في الصف نفسه من اللوحة نفسها. هذه الفجوة، وليست أي معيار أداء نشرته DeepSeek، هي ما يعنيه "تحسّن هائل في هندسة النوى".

لا تتوقف المقارنة عند صف الانتباه المتناثر، والصورة الأكمل تهم لأن عنوان التغريدة يذكر GLM-5.3 Flash بينما يشير المقال المكتوب المقتبس الذي تستشهد به فعليًا إلى طراز مختلف وأكبر هو GLM-5.2 في صف fused-MoE الخاص به — وهو عدم تطابق في التسمية يستحق الحذر بشأنه. في الصفوف التي أكمل فيها كلا الطرازين من فئة flash تشغيلًا فعليًا، يفوز V4.1 Flash في كل واحد منها: Kimi-Linear Decode عند 17.10× مقابل 13.64×، وMegaQwen Decode عند 5.39% من roofline مقابل 3.92%، وTopK Bitonic عند 4.25% مقابل 7.17% لـ GLM-5.3 Flash، وSonic MoE عند 10.81% (أفضل نتيجة سجّلها أي طراز على اللوحة لهذه النواة) مقابل 10.21%، وKDA CUTLASS عند 3.74% مقابل 2.26%. وفي نواتين إضافيتين لا يوجد لـ GLM-5.3 Flash أي تشغيل مكتمل على الإطلاق — fused MoE ونواة محاكاة Grid + MinGRU، حيث يسجّل V4.1 Flash نسبة 28.56% من roofline. (الصف الوحيد الذي يتفوق فيه GLM-5.3 Flash على V4.1 Flash هو TopK Bitonic، بنسبة 7.17% مقابل 4.25%، وهو أيضًا الصف الذي لا يقترب فيه أي من الطرازين من متصدر اللوحة.) الشكل العام مشابه لما في العنوان: في العمل منخفض المستوى على GPU، لا يتقدم V4.1 Flash على طراز Z.ai الأصغر فحسب — بل هو في مستوى مختلف تمامًا، وفي أصعب نواتين لا يكمل الطراز الأصغر على الإطلاق.

ما لا تفعله لوحة النتائج هو أن تخبرك بكم من تلك الفجوة يعود إلى النموذج وكم يعود إلى البنية التي يُطلب منه الكتابة لها. إن DeepSeek V4.1 Flash نموذج بـ552 مليار معامل، وآلية الانتباه الخاصة به قائمة على تصميم انتباه متناثر؛ أما GLM-5.3 Flash فهو نموذج أصغر لم يتمحور تدريبه قط حول حزمة CSA2 نفسها. والنموذج الذي يُطلب منه كتابة نواة لبنية تشبه توزيع تدريبه يمتلك ميزة متأصلة، ولهذا فإن أعدل قراءة للوحة هي قراءة اتجاهية لا حكم مواجهة مباشرة: فقد أصبحت هندسة النوى قدرة مقيسة لهذا النموذج، وهي قوية بحق، كما أن المقارنة المحددة التي أجراها صاحب التلميح تصح على الصفوف المشتركة. وتجيب اللوحة أيضًا عن سؤال لم تستطع ورقة المواصفات الإجابة عنه: هل يمكن تحقيق ذاكرة التخزين المؤقت بحجم 890 بايت عمليًا؟ والنموذج الذي يكتب نواة الانتباه المتناثر الخاصة به إلى نصف الـroofline هو نموذج يمكن، على الأقل، تنفيذ تصميمه الأساسي على عتاد حقيقي على يد نفس فئة المهندسين الذين سيتعين عليهم تشغيله في الخدمة.

الموضع الصادق الذي يستقر فيه هذا الأمر هو بجانب كل شيء آخر في هذا المقال: قياس آخر من طرف ثالث لم تنتجه DeepSeek، وهذه المرة لقدرة — كتابة نواة GPU — لم يغطّها أي صف في اختبارات البائعين المعيارية. إنها لوحة مجتمعية واحدة على جيل GPU واحد، أُجريت مرة واحدة، وجولة التدقيق الخاصة بها ليست محكّمة من الأقران، لذا فهي دليل على الاتجاه لا نتيجة معتمدة. لكنه أول رقم خارجي يُظهر الغرض من البنية الجديدة: نموذج Flash-tier رخيص، عندما يُطلب منه أداء العمل على الأنظمة منخفضة المستوى الذي تنفق DeepSeek نفسها هندستها عليه، يبلغ نحو نصف ما يسمح به العتاد، بينما يبلغ منافسه المباشر من فئة Flash-tier أقل من واحد على عشرين.

إصلاحات سريعة، تأكيدات بطيئة: أولى التقارير العملية

الإنتاجية هي النصف الذي لا يستطيع عدّاد الرموز رؤيته من قصة السرعة. أما التقارير العملية التي تراكمت منذ الإطلاق، فهي تقيس النصف الآخر — كيف يتصرف النموذج وهو في طريقه إلى إجابة، وما تكلفة هذا السلوك — وهي لا تتفق على ما إذا كان ما وجدته ميزة. تصف ثلاثة روايات مستقلة النمط نفسه. ولا شيء منها تقييم مضبوط؛ فالأرقام الواردة أدناه تخص ممارسين أفرادًا يشغّلون مهامهم الخاصة، وينبغي قراءتها كتقارير استخدام لا كقياسات.

أوضح وصف لهذا السلوك يأتي من MindStudio، التي وضعت DeepSeek V4.1 Flash في مهمة بناء عارض ثلاثي الأبعاد تفاعلي من الصفر. سجل استدلاله، بحسب كلمات المراجع، «أظهر أنه يتحقق من حساباته الخاصة بدلاً من مجرد إصدار الكود». التقط النموذج خطأً في تخميد التحكم المداري أثناء البناء، واستنتج التصحيح من المبادئ الأولى، ثم «أعاد التحقق من الإصلاح في متصفح بلا واجهة قبل أن يعلن انتهاء المهمة». وفي مهمة ثانية — لوحة معلومات قائمة على Docker — وجد معامل مقارنة واحدًا مقلوبًا وأكّد الإصلاح عبر استطلاع نقطة النهاية الحية ثماني مرات بينما كانت الأرقام تتحدث. هذا ما كان يصفه مختبرو النسخة التجريبية عندما سموا النموذج مفرطًا في التفكير، ولا خطأ في أي خطوة واحدة من ذلك: فقد وُجد الخطأ وأُصلح وتم التحقق منه. وهو أيضًا بالضبط مصدر الشكوى. فقد كان الخطأ قد عولج بالفعل، وكانت جلسة DeepSeek أخرى قد أكدت الإصلاح خلال دقيقة، ومع ذلك واصل النموذج التحقق. يرى MindStudio في هذه العادة قوة ويقول ذلك صراحةً؛ وكان انتقاده الوحيد في ذلك الاختبار مشكلة فيزيائية دخلت في حلقة حتى أُوقفت يدويًا في المحاولة الأولى، ثم حُلّت بشكل صحيح عند إعادة المحاولة.

القراءة من الاتجاه الآخر تدور حول الساعة. يصف تجميع مجتمعي لاختبارات V4.1 Flash مسألة فيزيائية — كوب ماء مقطوع بزاوية مائلة — حيث قضى النموذج اثنتي عشرة دقيقة وحوالي 130,000 رمزًا في طريقه إلى إجابة صحيحة، وأجرى ثلاث جولات من التحقق الذاتي لم يُطلب منه إجراؤها قبل أن يتقارب. يوثّق نفس التقرير نمط الفشل عند الطرف الأقصى لتلك العادة: تقع سلسلة التفكير أحيانًا في حلقة تكرار "دعني أكتب" التي تدور عبر الرموز بسرعة 200 في الثانية تقريبًا دون تقدم، ونصيحته تتعلق بجانب العميل لا بجانب النموذج — ضع سقفًا لـ max_tokens ومهلة للطلب ودعهما يقطعان الحلقة. كما يسجل أن نفس فئة المسألة تم إنجازها في ثماني عشرة ثانية بواسطة تشغيل مع خفض جهد الاستدلال إلى منخفض، وهو تذكير بأن معظم هذا قابل للضبط وليس عيبًا.

أكثر الأدلة حِدّة على أن حجم التحقق ليس مرادفًا للصواب يأتي من مراجعة أجرتها Tencent Cloud Labs، قُرن فيها DeepSeek V4.1 Flash كمنفّذ مع GPT-6 Astra كمراجع في إعادة إنتاج ورقة بحثية. أُجريت ثلاث جولات، ولم تنقض Astra الخوارزمية الأساسية قط. وما ظل المراجع يصححه، جولة بعد جولة، هو البنية المساندة التي بناها V4.1 Flash لنفسه — إعداده التجريبي، وبنائه الإحصائي، والتفسير الذي استخلصه من نتائجه الخاصة. نما عدد الاختبارات من 82 إلى 97 إلى 110، ونجحت جميعها في كل مرة. وبالمقارنة مع خط أساس بالقوة الغاشمة، طابقت التهيئة التي اختارها النموذج التهيئة المثلى في 8.89% من الحالات وفق معاملات الورقة نفسها، و11.04% مع ثقة عشوائية، و10.30% ضمن تهيئة عشوائية. وكان أكثر ما توصلت إليه Astra حِدّةً يتعلق باختبار ادّعى أنه يتحقق من حدود الأعداد الصحيحة: فقد نفّذ 153 فحصًا امتداديًا وصفر فروع لحدود الأعداد الصحيحة. يمكن لنموذج أن ينجح في مجموعة الاختبارات التي كتبها ومع ذلك ألّا يكون قد أجاب عن السؤال، وتلك المراجعة حالة عملية لهذه الفجوة بدلًا من مجرد تأكيد بوجودها. النطاق ضيق، ويقول المؤلفون ذلك — محدد واحد لطول تخميني، وقيم ثقة محاكاة، ولا تشغيل من طرف إلى طرف — لكن الشكل هو ذاته الذي تصفه التقارير الأخرى.

عند وضعها إلى جانب أرقام Artificial Analysis، يتوقف هذا عن أن يبدو حكاية عابرة. قيّم أول تقييم مستقل للنموذج بأنه سريع جدًا ومسهب جدًا في آن واحد: 185 رمز إخراج في الثانية، و250 مليون رمز إخراج وُلّدت عبر مجموعة تقييمه مقابل وسيط قدره 140 مليون للنماذج المماثلة. لا تستطيع درجة مركبة أن تقول أين تذهب المخرجات الإضافية. لكن التقارير العملية تستطيع. هذه الرموز ليست حشوًا — بل هي النموذج وهو يتحقق. ويظهر الحجم نفسه من اتجاه غير ذي صلة في تقرير خطأ عن عميل طرفي تابع لجهة خارجية، حيث أُلحق محتوى الاستدلال على مسار deepseek-flash بالإجابة المرئية بدل فصله في كتلة تفكير، ما أنتج كتلًا نصية مفردة تصل إلى 73,003 حرف من الحوار الداخلي، دون التقاط أي شيء لإعادة تشغيله في الأدوار اللاحقة. هذا الخطأ يخص العميل وليس DeepSeek، وهو ليس دليلًا على الجودة — لكنه توضيح منصف لمقدار التفكير الذي يصدره هذا النموذج عندما يفشل إطار التشغيل في فصله.

لا شيء من هذا يقلب حسابات السعر أعلاه؛ بل يفسّرها. النموذج الذي يتحقق من نفسه يستحق ما يستهلكه من رموز في الأعمال الوكيلية، حيث تكلفة الإصلاح الخاطئ أكبر من الرموز التي تُنفق لإثبات صحته، وهو غير مناسب لحلقة طويلة من التعديلات التافهة يكون فيها التأكيد هو معظم التشغيل. هناك أمران يجعلان الفرق قابلاً للإدارة. مقبض جهد الاستدلال عدد صحيح متصل من 1 إلى 100 بدلاً من ثلاث طبقات ثابتة، فيمكن خفض النموذج نفسه للأعمال الرخيصة ورفعه للأعمال الصعبة — والنتيجة البالغة ثماني عشرة ثانية أعلاه جاءت من فعل ذلك بالضبط. وشبكة الأمان الرخيصة تنتمي إلى العميل: سقف max_tokens ومهلة الطلب يحوّلان حلقة التكرار من فاتورة إلى استجابة مقتطعة. وكلاهما من الإعدادات التي تستحق التثبيت قبل أن يوجّه مسار إنتاجي نظره إلى هذا النموذج، ولا يفعل النموذج أياً منهما من تلقاء نفسه.

حالات الرفض: ترفض نقطة التحقق، والرفض قابل للقياس

On October 20 the thing drawing attention to DeepSeek-V4.1-Flash-1017 was not a benchmark row but a refusal. The DeepSeek tracker this piece has already quoted posted a screenshot from an agent session running the 1017 checkpoint with dsh-purge applied, and reported that the model declined anyway: the run, in the poster's words, “refused the persona,” with the model reasoning that the request was “a prompt injection attempt” and that it “should not adopt the persona.” The post's reading was that the checkpoint “is *not* entirely naive about jailbreaks” and that DeepSeek is “starting to think about safety.” dsh-purge is not a prompt trick pasted into a user message. It is a public DeepSeek Harness plugin — 9,897 stars on GitHub and version 1.1.71 as of October 20 — built to strip the harness's own prompt rewriting and injected suffixes, and its English README draws exactly the distinction this test turns on: the harness “rewrites prompts, approval gates, sandbox policy, and session migration at runtime,” so asking only “will the model answer” treats a host-layer block as a model refusal, and the plugin exists to split those two layers on your own install. In this session the separation came out on the model's side. The counters in the same screenshot — 156 steps, 33.7 million tokens, 232 tokens per second — describe a full agent run rather than a one-shot prompt, and the agent's own summary says the plugin was mechanically fine and “that specific model declined it.”

That is one person's screenshot of one session, and it should be carried as exactly that. The prompt is not shown, the verdict is the agent's own summary rather than a graded response, and the account is a DeepSeek watcher rather than an evaluation lab, so nothing in the post can be reproduced from the post. What makes it worth taking seriously is that both halves of it can be checked against sources independent of the person who posted it: the plugin, which is public and documents its own mechanism, and the checkpoint's refusal behaviour, which somebody else has now counted.

The count is in the paperwork of a derivative. A community abliterated build of the 1017 weights, published on Hugging Face on October 17, carries the measurement that made the build necessary. On 100 held-out harmful requests, decoded greedily for 384 tokens each, the untouched checkpoint refused 97 in chat mode and 98 in think mode; after a rank-2 refusal subspace was projected out of the 40 main layers, the same 100 prompts drew 4 and 9 refusals. Those are third-party figures about DeepSeek's weights, produced without access to DeepSeek's training pipeline, and they are also one prompt set, one grader and one quantisation of the model — a Q8_0 conversion rather than the FP8 DeepSeek ships — so the honest reading is “almost always,” not “97.0%.” On that evidence the poster's description holds: the checkpoint refuses, reliably, in both of its modes.

ما لا يدعمه الشكل هو فكرة أن هذا موقف جديد. كانت نقطة تحقق سبتمبر موجودة بالفعل. وقد قاس الفريق الذي يقف وراء نسخة سبتمبر غير الخاضعة للرقابة النموذج غير المعدّل على مجموعة HarmBench الكاملة المؤلفة من 320 مطالبة، وأبلغ عن نجاح الهجوم بنسبة 42.8% مع إيقاف تشغيل الاستدلال، و1.6% عند أقصى جهد — أي إن النموذج الذي يحل هذا محلّه رفض نحو 98% من تلك الفئة من الطلبات متى سُمح له بالتفكير. يأتي القياسان من مجموعات مطالبات مختلفة ومقيّمين مختلفين وتكميمات مختلفة، لذا لا يمكن طرح أحدهما من الآخر؛ وما يتفقان عليه هو الشكل العام. كان الرفض عند الجهد العالي سلوك نموذج سبتمبر بالفعل، وتقع نقطة التحقق 1017 في النطاق نفسه. أما الادعاء الأضيق فهو ما تختبره جلسة 20 أكتوبر فعليًا، وهو الجدير بالاحتفاظ: فقد صمد الرفض أمام إضافة بُنيت خصيصًا لإزالة القيود على جانب الإطار، وصمد بوصفه حكم النموذج على المطالبة لا بوصفه حظرًا من طبقة المضيف.

None of this is documented by the vendor. DeepSeek's September model card has an evaluation section and it is benchmarks only — no refusal rate, no red-team result, no safety evaluation — and the 1017 card carries no evaluation section at all, having dropped that folder along with the technical report. Refusal behaviour for this model is therefore described entirely by the people building around it: derivative authors counting prompts, and a tracker reading a transcript. For anyone pointing an agent at the checkpoint, that is the operative gap. A refusal inside an agent loop is not a content-policy footnote; it is a failed tool call in the middle of a run, and it surfaces like any other per-call failure — which is why the mitigation is the one a timeout gets. On OrcaRouter that is a routing decision rather than a code change: the 1017 checkpoint and the DeepSeek models it would fall back to sit behind one endpoint, so a call that comes back refused can be retried against a second model instead of ending the run. The refusal itself belongs to the model, and nothing on the routing layer changes it.

عدد النسخ المشتقة هو آخر عنصر مستقل، وهو يشير في الاتجاه نفسه. أظهر بحث على Hugging Face باسم نقطة الحفظ 121 مستودعًا عامًا في 20 أكتوبر، و25 منها تحمل في اسمها كلمات غير خاضعة للرقابة أو abliterated أو derisked — إصدارات هدفها كله إزالة السلوك الذي صادفته جلسة 20 أكتوبر، ونُشر أولها في اليوم نفسه الذي رُفعت فيه الأوزان. وهذا ليس دليلًا على جودة النموذج، وليس توصية؛ بل هو دليل على وجود آلية رفض عاملة في الأوزان يمكن إزالتها، لأن من يزيلون هذه الأشياء لا يتكلفون عناء ذلك حين لا يكون هناك شيء يُزال. كما يحدد التوقع لأي شخص يستضيف النموذج ذاتيًا: نقطة الحفظ التي تنزّلها هي التي ترفض، وكل نسخة غير خاضعة للرقابة في تلك القائمة هي نموذج مختلف، وليست إعدادًا.

سؤال V4 Pro، والدرجات فوق Flash

During the beta, DeepSeek's feedback form asked testers whether the intermediate version "can fully replace the online DeepSeek V4 Pro." That was a striking question to put to users, because it implied the flagship's future was open. The answer that arrived on September 9 and 10 was an operational one: from 12:00 Beijing time on September 14, deepseek-v4-pro would stop being served and its traffic would land on V4.1 Flash at Flash prices, with V4.1 Pro named as the eventual successor. Developers objected, and the objection was not that the replacement was bad — it was that a model string is a contract, and swapping the weights behind a production ID breaks prompt-tuned pipelines and agent loops built against the old ones. Within two days DeepSeek withdrew the plan. Its own changelog now says it has “decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method remaining unchanged,” and that it will give further notice if that changes. The reversal is vendor-stated rather than reported, and it is checkable on the Models & Pricing page, which still lists deepseek-v4-pro beside deepseek-flash.

What the routing table does not explain is why. On September 9, teortaxesTex — one of the tipster accounts that flagged the beta — reported that DeepSeek had conceded its V4-Pro pre-training went wrong; in his words, the company “dun goofed” with it. That remains a single-source claim from an insider-leaning account, and DeepSeek has conceded nothing publicly: the changelog frames the phase-out as a consequence of the new model being better, not of the old one being broken. What can be checked from outside is the shape of the August 13 V4 Pro general-availability launch. The announcement was pulled within about a day. The Hugging Face repository briefly carried a Flash-tier configuration under the Pro name before the weights were corrected. And independent scoring of what was being served under that name landed barely above the much smaller DeepSeek V4 Flash, against official charts showing near-flagship results — on the current artificial-analysis index the two sit a single point apart, 36 against 35. Independent write-ups have blamed different causes — reasoning behaviour overfit to DeepSeek's own evaluation harness, a deployment and weights mixup, or, as the leak claims, a pre-training that genuinely misfired. The rerouting does not settle which, and DeepSeek has since called it off; what survives is a vendor pricing its newest model at a quarter of the flagship's cache-miss rate and putting it ahead of that flagship on its own benchmarks, which companies do not usually do because everything went to plan.

Flash, Pro, Ultra: what DeepSeek has said, and what is only a guess

DeepSeek's own documentation is explicit that this is the bottom of something. The changelog calls DeepSeek V4.1 Flash “the smallest model in our new architecture family” and says the architecture “is designed for a higher capability ceiling, faster inference, higher throughput, and scaling to larger models.” The technical report closes the same loop in stronger language: the model “serves as a new starting point for our continued scaling efforts,” and the company says it “will pursue the joint scaling of model architecture, pre-training, and post-training to further explore the frontier of model intelligence.” Both sentences are first-party, both were published on September 10, and neither is a product announcement. They commit DeepSeek to building bigger models on this design; they do not name one, size one or date one.

Everything past that line is speculation. On September 25 the tracker @teortaxesTex — the same account whose September 9 post on the V4 Pro pre-training this piece already carries — read the pattern as a full three-tier line: “DeepSeek is pretty clear that they're not going to stop here… This sounds like they intend to make a whole lineup, Flash-Pro-Ultra. Maybe 522B-2T-8T?” The sizes are labelled a guess by the person offering them, and the hedge is doing real work: nothing in DeepSeek's changelog, model card, technical report, Hugging Face organisation, pricing page or model listing names an Ultra tier, and nothing above 552B. The guessed ladder is also shakier arithmetic than it looks. Its bottom rung, 522B, is not the 552B model that shipped — either the figure is a slip for a number DeepSeek has already published, or it points at a checkpoint nobody has seen, and there is no way from outside to tell which. Its top rung, 8T, is five times the 1.6T flagship whose repository DeepSeek opened in April, in a year when the company's most recent release was engineered to activate 8B parameters per token because inference cost, not parameter count, was the constraint it set out to solve.

The same account has separately floated a 3-trillion-parameter DeepSeek model as something the lab was working on but was unsure it could serve economically — a different claim about a different thing, and equally unconfirmed. Neither has to be reconciled with the other, because neither has a second source. What can be checked is the direction: a lab that calls its newest model the smallest in a family, and its architecture one that scales to larger models, is a lab planning to ship above this rung. What it will call the rungs, and how big they will be, is not in any document a reader can open today.

For anyone choosing a model rather than following a roadmap, that distinction is the whole of it. If the family does extend upward, the decision in front of most teams is still whether to move onto the rung that exists — and the cheapest way to hold that decision open is not to wait. DeepSeek V4.1 Flash is on OrcaRouter at DeepSeek's own rates with 0% markup, so if a Pro-tier successor arrives at a different price, the change lands in the rate card rather than in a contract. DeepSeek V4 Pro sits on the same key behind it, which makes a third tier arriving later a routing decision rather than a migration.

أوزان MIT تغيّر ما يمكنك فعله به

DeepSeek V4.1 Flash ships under the MIT licence, and the weights went up in the deepseek-ai organisation on Hugging Face on September 10 — the same day as the API, not weeks later. The repository is deepseek-ai/DeepSeek-V4.1-Flash: 48 safetensors shards totalling about 510 GB, the technical report as a PDF beside them, and an inference folder holding a reference implementation plus a conversion script that splits the checkpoint across tensor-parallel ranks. That repository is the September checkpoint; the live one since October 17 is deepseek-ai/DeepSeek-V4.1-Flash-1017, which carries the same MIT licence, the same 48 shards and the same roughly 510 GB, but which replaces every shard and drops the report, the chat template and the evaluation folder.

من المفيد أن نكون دقيقين بشأن ماهية ذلك التنزيل، لأنه ليس إصدارًا كامل الدقة. تحتوي ملفات safetensors على نحو 204B معلمة بصيغة FP8 و557B بصيغة INT8 مقابل نحو 2B بصيغة BF16، ويصرّح ملف الإعداد بتكميم fp8 — فنقطة التحقق التي تسحبها مضغوطة مسبقًا، وبهذه الطريقة يتسع نموذج بـ763B معلمة في نحو 510 GB. أكبر كتلة مفردة داخله هي جدول Engram: جدولان من FP8 يحويان نحو 189 GiB من التنزيل، ولا تقرأ منهما كل خطوة فك ترميز سوى بضعة صفوف.

بالنسبة إلى نموذج بعمود فقري بحجم 552B، يكون الترخيص هو الجزء الذي يغيّر القرارات. فرخصة MIT متساهلة، بلا بوابة عائدات وبلا بند إضافي للاستخدام المقبول يستحق مراجعة قانونية، أما ملف التنشيط — 8B لكل رمز عند التعبئة المسبقة، و16B عند فك الترميز، مع ذاكرة تخزين مؤقت بسعة 890 بايت لكل رمز — فهو ما يجعل تقديم الخدمة به ممكناً خارج مزوّد سحابي فائق، حتى إن بدا عدد المعاملات كبيراً. لكن هذا الرقم الأخير خاصية لمكدس التخزين المؤقت الكامل لا للأوزان، لذا فإن محرّكاً مستضافاً ذاتياً ينفّذ الصيغة المُكمَّمة دون بقية التصميم سيستقر أعلى منه بكثير.

A permissive licence is not an export clearance, and for six days in October the difference looked as though it might matter. On October 16 China's Ministry of Commerce and Ministry of Science and Technology published Announcement No. 48 of 2026, a draft revision of the catalogue of technologies prohibited or restricted from export that reaches AI model weights and source code for the first time. It proposes 40 additions across AI, biotech and solar, 23 of them AI-related, with public comment running to November 15. Two carve-outs in the draft do most of the work for a model like this one: technology already in the public domain through open-source release or academic publication is not to be controlled, and API, app, SaaS and lightweight on-device access for overseas users is not treated as an export of model technology. The thresholds reported across the trade press combine a parameter floor of two trillion, a place in the global top ten on performance and safety benchmarks, and training on core or important national data. A 552B model clears the first; whether the other two catch it is not something a reader can determine from outside, because the draft does not publish the definitions — which makes the public-domain carve-out, not the size test, the clause that matters for weights already on Hugging Face.

What changed on October 22 was not the text but the compliance record. YuyuantanTan reported that DeepSeek and MiniMax had completed the registration procedures and finished open-sourcing and overseas licensing. The report does not say which releases it counts as the open-sourcing, it names no model, checkpoint or date, and it announced no new weights; the open-source releases visible in that window are the DeepInfer engine and the deepseek-recipe toolkit covered above. The build-level version of the claim is narrower and later: on October 23 an X account with a long record of tracking DeepSeek posted that DeepSeek and MiniMax had “obtained permissions for open sourcing V4.1F-1017 and M3.1-Flash” respectively. That attribution is the poster's, neither company has published a statement confirming it, and no regulator has published a notice naming either company's registration. The checkpoint the post names had been on Hugging Face under MIT since October 17, six days earlier. Whatever permission was granted, it did not gate the download: the repository came first.

For a reader, the practical content is narrower than the headlines around it, and it runs in two directions. Nothing about the October 17 checkpoint, the MIT licence or the repository changed on October 22; the download, the shards and the hash differences described above are exactly as they were. What changed is the answer to whether a Chinese lab can keep publishing weights like these while the catalogue is still in consultation: the state-media account says yes for DeepSeek and MiniMax, under a draft that excludes already-public technology and leaves API access outside the export definition. The limit is in the tense. The exemption is for technology already in the public domain, which describes the September and October checkpoints and would not automatically describe a release made after the catalogue is finalised — and the interim registration duty said to have taken effect on October 16 has been reported rather than published in full. Anyone planning to download whatever DeepSeek ships next is planning against a rule that does not exist yet. Anyone calling V4.1 Flash today, or running the 1017 weights, is relying on a carve-out that the draft states and that neither account of the interim measure contradicts.

Serving support arrived the same day, and it sets the practical floor. vLLM published a day-0 recipe validated on NVIDIA and AMD hardware, and SGLang shipped day-0 inference and RL support in the same window through its Miles training stack. Both shipped as preview images rather than stable packages — the vLLM container carried a September 9 tag, a day ahead of the model's own GA. That side of the story moved on September 22, when vLLM released v0.30.0: the first stable vLLM version to carry this architecture, with DeepSeek V4.1 Flash listed under its new-model support, the model's own vLLM recipe now requiring 0.30.0 or later, and the launch-day image tag retired. The release notes are the project's own account of its work rather than a measured result, and should be read that way. They credit the version with holding the model's whole KV cache in MXFP8 through the FlashMLA V4.1 record on SM100, and with asynchronous prefetch for CPU-offloaded Engram lookups plus Engram data-parallel sharding — the two changes that matter most for a model whose Engram table is the single largest block in the download. Neither of those two changes carries a number outside the project, and the release notes attach none — but the project published a before-and-after on the model as a whole on October 7, three weeks after release: 1.9× faster at low concurrency and 5.3× the throughput under a 150 TPS constraint against its own day-0 build, with the merged pull request behind each optimisation listed. The figures and their limits are in the speed section above. What does not exist is an outside replication — every number in that post is the project measuring itself. What the release does settle is packaging: 762 commits from 315 contributors, 104 of them first-time, in a tagged version rather than a preview image. The vLLM recipe's own arithmetic still puts the minimum at roughly 614 GB of VRAM for the 510 GB checkpoint plus headroom — an eight-GPU H200 node or a GB200 NVL4 tray — and warns that a 1M-token context wants to be capped in practice rather than assumed. A narrower KV cache changes how much memory the running engine needs, not how large the checkpoint is, so the node floor stands. SGLang's write-up adds a lever specific to this architecture: moving the Engram table into host memory with a single environment flag raised KV cache capacity by about 36% on a four-GB300 test at comparable decode throughput and time to first token, and its decoder-side SWA bounded replay lifted prefill throughput 1.56× on eight H200s. Two limits travel with that. The reference implementation in the repository describes itself as a readable example rather than a production serving engine, and the consumer runtimes lagged: the quantization path that would put this on a laptop or a Mac Studio was not open at launch. The short version is that the weights are genuinely open and the serving stack genuinely exists — on hardware counted in nodes rather than GPUs, which is the honest reason a permissive licence does not automatically mean a local build. The 1017 repository changes nothing about that floor — the checkpoint is the same 510 GB across the same 48 shards, and the runtime notes that carry it are the ones published for September's weights.

What open weights do not give you, on their own, is a quality signal — but that gap has now partly closed, and it closed further on October 18. Four independent instruments now exist, one of them run twice: Artificial Analysis's Intelligence Index, which scores the September checkpoint at 39 and the 1017 checkpoint at 45; the Exploit Bench study above, a cyber benchmark run under seven agent harnesses; the community kernel board in the section below; and, since October 14, Artificial Analysis's own AA-AgentPerf results for this model, which measure the hardware it is served on rather than how good it is. The vLLM project's October 7 before-and-after, above, measures a different thing again — what the model costs to serve rather than how good it is — and is the runtime's own account of itself. None is a reasoning score of the kind ARC Prize is building, three of the four still describe the September checkpoint, and for most of the specific rows in this piece the model card remains the only source — but the composite no longer belongs on that list: the 45, and the 33.8 under it, were produced against the weights DeepSeek serves today. That is not an accusation; it is the state of the evidence, and it is why this article keeps the two kinds of number apart. Open weights do mean the model will not stay confined to DeepSeek's endpoint, which matters for anyone whose blocker was data residency rather than price.

كيف تتبنّاه دون المخاطرة بالإنتاج عليه

One thing changed on our side since the release, and it makes this section simpler than it would have been last week: DeepSeek V4.1 Flash is on OrcaRouter now. It answers to the model id deepseek/deepseek-v4.1-flash through the same OpenAI-compatible endpoint as the rest of the line, at DeepSeek's own rate — $0.15 per 1M input tokens, $0.60 per 1M output and $0.003 per 1M cache reads at off-peak rates, doubling inside the weekday peak windows — passed through with 0% markup. That last part is the one worth spelling out: the figure you are billed is the provider's figure rather than a resale price, so a vendor price cut is live here the same day instead of whenever a contract is renegotiated. DeepSeek V4 Flash and DeepSeek V4 Pro sit on the same key, which makes the migration below a routing change rather than a re-platforming. The October 17 checkpoint moves none of that: DeepSeek kept both the API model name and the rate card, so the update changes neither the id nor the price on this side, which is what a pass-through looks like when a vendor refreshes a model without touching either.

That is also the argument for not rushing the migration. The September 14 rerouting would have been a vendor changing what a model ID means underneath you, had DeepSeek not cancelled it under pressure, and the episode is the cleanest recent reminder that a model string is a contract a provider can rewrite — which is also the right frame for the line-up talk above, since a three-tier family is a plan and this one has already changed once. For a model whose only outside scores so far are a composite index, a noisy cyber table and a community kernel board — and which has no production record at all — the safer shape is the one a router is built for: send a slice of production-shaped traffic to the new model, hold a proven model behind it as an automatic failover so a quality regression costs a retry instead of an incident, and let the routing DSL decide which calls are worth the unproven model and which should stay on the workhorse. The migration to V4.1 Flash is probably the right destination. It is not a migration to run on the announcement.

أسئلة مفتوحة

• Independent benchmarks — the count has gone from zero to four instruments, and none of them measures the model's reasoning the way ARC Prize intends to. Artificial Analysis's score of 39 is a composite of ten evaluations at maximum reasoning effort, and its 45 — published October 18 — is the same composite pointed at the newer weights. The Exploit Bench study above is a cyber-exploitation score inside a fixed token budget, and it rates DeepSeek V4.1 Flash below GLM-5.3-Flash. The community kernel board on kernelbench.com is a kernel-engineering table. Artificial Analysis's AA-AgentPerf results, published October 14, are a hardware table: how many tokens per second per megawatt each accelerator system serves this model at, and nothing about quality. Nobody has replicated DeepSeek's individual benchmark rows, and no independent measurement at production effort settings, or at the far end of a 1M-token window, has been published. The October 17 checkpoint is no longer unmeasured: on October 18 Artificial Analysis added a 1017 row at 45 on its index and 33.8 on Terminal-Bench 4.0, and the bench figure sits eight and a half points under what the vendor table claims for the same weights. What is missing now is a second opinion on that figure — one evaluator, one harness, one run — and any independent score at all for the four measurements that predate the swap. The one evaluation built to answer the reasoning question is still under construction — ARC Prize has a DeepSeek V4.1 Flash profile in its benchmarking harness in an open pull request, with no result attached — and ARC-AGI has never scored this checkpoint at all: the 61.4% on ARC-AGI-2 that the adapter's arrival revived belongs to DeepSeek V4 Flash 0731, the model V4.1 Flash retires. The newest number attached to this model, the 77.0 on DeepSWE v1.1, is the weakest-sourced row in the piece: it is DeepSeek's own chart, read off an image attached to the vendor's announcement, and no outside run of DeepSWE v1.1 on the 1017 checkpoint exists — Artificial Analysis's October 18 result is for the same weights but for a different benchmark. The two independent boards that publish the benchmark put this model's name at 71.7% and 74.2%, which is enough to place it beside Opus 5 and not enough to confirm the vendor's figure — and neither of them says which checkpoint it measured.

• Self-hosting below the node floor — vLLM's day-0 support became a stable release on September 22 with v0.30.0, which adds an MXFP8 KV path and asynchronous Engram offload; SGLang's day-0 support still ships as a preview image. Neither moves the hardware floor, which sits around an eight-GPU node because the 510 GB checkpoint is what sets it — and DeepSeek's own deployment notes, published with DeepInfer, put a finer point on it: one prefill node beside one decode node is deployable, but the recommended decode configuration is expert parallelism of 64 or above, which makes the smallest efficient unit 64 GPUs of prefill beside 64 of decode. The vLLM project's October 7 post is the fullest measured optimisation ledger published on this model — 1.9× at low concurrency and 5.3× under a 150 TPS constraint against its own day-0 build, alongside the SGLang figures above — and its October 17 follow-up adds a tokens-per-dollar curve that does span H100, MI300X and B200 configurations beside the rack-scale parts. Both are the project's own accounting of its own runtime. Neither moves the floor: a cheaper token on a GB300 or an H100 still does not fit a 510 GB checkpoint onto a desktop, so it is a faster path to the same floor rather than a lower one. Whether later releases and community quantizations bring this onto smaller machines is the open question for anyone who wants the weights rather than the endpoint, and it is the one that decides whether a permissive licence is a practical licence.

• ما هو V4.1 Pro — تسميه DeepSeek خليفةً لـ V4 Pro، لكنها لم تنشر أي تاريخ، ولا مواصفات، ولا سعر، وقد سُحب الإعلان عن إعادة توجيه حركة Pro الحالية إلى طراز من فئة Flash قبل أن يدخل حيز التنفيذ. أما ما يقع فوق Flash، فلا تصفه الشركة إلا بأنه المعمارية التي تتوسع إلى نماذج أكبر؛ أما سلم Flash وPro وUltra وأحجامه 522B و2T و8T فهي تكهنات أحد المراقبين، وليست من DeepSeek.

• Per-task cost — Artificial Analysis now carries a per-task figure for both checkpoints: $0.49 per Intelligence Index task for the September weights, restated from the $0.27 this piece first carried, and $0.26 for 1017. That is the first outside evidence for any part of DeepSeek's Pareto claim, and it is narrower than the claim. The same rows show the 1017 checkpoint generating 210M output tokens across the index against 250M for September — shorter outputs in aggregate — but 89,495 output tokens per task against 88,574, which is not shorter at all, and 288.8 seconds per task against 289.9, which is not faster. The site's own cost breakdown puts the whole of the saving on the input side: the output line is flat, $0.106 per task against $0.107, while input falls from $0.379 to $0.155 and cache writes from $0.322 to $0.095. Whether a task finishes on fewer dollars depends on how many tokens it takes, and that is a property of the workload rather than of the rate card. DeepSeek's own account of the update points the other way: the changelog claims the new checkpoint reduces task runtime and output length, and its October 17 post repeats it as “faster task completion, and shorter outputs.” Neither carries a number, and the outside measurement behind this bullet has now met them: on the two quantities DeepSeek named, the 1017 weights moved by about one per cent, in opposite directions, from a starting point of five weeks earlier. A third figure sits on the other side of the same ledger: the vLLM project's October 17 chart puts the model at 1.3B tokens per dollar of total cost of ownership on SemiAnalysis AgentX, at a cost tier that assumes large-hyperscaler volume. Cheaper serving does not settle this bullet either — a lower cost per token leaves the tokens-per-task figure, which is what a caller actually pays for, exactly where the September measurement found it.

• التحقق الجامح — يُبلِغ المختبِرون الميدانيون عن حلقات تكرار عارضة في سلسلة التفكير، وعن مداولات تستغرق اثنتي عشرة دقيقة وتستهلك مئات الآلاف من الرموز في مسائل تُحل في ثوانٍ عند إعداد جهد أقل. ولا يقيس أي معيار مستقل وتيرة حدوث أي منهما ولا تكلفته في بيئة الإنتاج، كما أن التخفيف المُبلَّغ عنه حتى الآن هو حدود قصوى للرموز من جانب العميل، لا شيء يفعله النموذج من تلقاء نفسه. وتوثّق دراسة Exploit Bench أعلاه نمط الفشل نفسه من الاتجاه المعاكس: فحلقة التقرير المتكرر التي تُظهرها سجلات ReAct ينتجها موجّه الضغط في منصة التشغيل، لا استدلال النموذج نفسه، وهو تذكير بأن وجود حلقة في سجل ما ليس عيبًا في النموذج تلقائيًا.

• هندسة النوى — لوحة صدارة نوى CUDA المجتمعية الخاصة بـ kernelbench.com ‏(RTX PRO 6000، مقيّسة وفق roofline) تضع DeepSeek V4.1 Flash عند 50.19% من roofline المكافئ للكثيف على نواة الانتباه المتناثر الخاصة به، في المرتبة الرابعة إجمالًا، مقابل 4.45% لـ GLM-5.3 Flash في الصف نفسه. اللوحة يديرها المجتمع وليست خاضعة لمراجعة الأقران؛ ولم يُنشر جيل ثانٍ من GPU أو تكرار للنتائج.

• السياق الطويل في الطرف البعيد — دُرِّب الانتباه المتناثر عند 64K ووُسِّع إلى 1M عند علامة 34T رمزًا. إن كيفية صمود الجودة قرب نهاية نافذة كاملة بطول 1M هي بالتحديد ما سيحسمه تقييم مستقل للسياق الطويل.

• Cache reconstruction at the edges — SWA Bounded Replay rebuilds dropped sliding-window state approximately rather than exactly, and DeepSeek says the result can vary depending on where a cache hit lands in a session. The report notes little observed quality impact; extreme or untested conditions are exactly where an approximation like that would surface, and nobody outside DeepSeek has looked.

• ما إذا كانت أعباء عمل Pro تنتقل فعليًا — فقد سُحبت إعادة التوجيه القسرية، لذا لا شيء ينتقل ما لم يختر فريق أن ينتقل، والفرق التي دفعت أسعار Pro مقابل عمق الاستدلال هي الأكثر ترجيحًا لملاحظة فرق في القدرات في العمل الشبيه بـ Terminal-Bench 3.0، حيث يتخلف V4.1 Flash عن Opus-5.0 بفارق كبير.

• شروط الطبقة المجانية وسلوكها — أضافت OpenCode إدخالًا مجانيًا لهذا النموذج في 15 أكتوبر، على نافذة قدرها 200,000 رمز وسقف إخراج قدره 32,000 رمز، وُصِف في صفحاتها الخاصة بأنه متاح لفترة محدودة. أما ما إذا كان هذا الترتيب سيبقى بعد تلك العبارة، وكيف تصمد نقطة النهاية تحت الحمل، وما إذا كان لـ DeepSeek أي رأي في ذلك، فكل ذلك غير معلن: قائمة النماذج الخاصة بالمنصة ووثائقها وجدول أسعارها هي المصدر الوحيد، ولا يوجد تاريخ انتهاء لإلزام أي أحد به، ولم ينشر أحد قياسًا للمسار المجاني.

• Safety and refusals — DeepSeek publishes no refusal rate, red-team result or safety evaluation for either checkpoint, and no independent lab has scored this model on a jailbreak benchmark. The figures in this piece come from people building uncensored derivatives — 97 of 100 harmful prompts refused in chat mode, 98 in think mode, on one prompt set and one grader — and from a single agent-session screenshot in which the model declined a jailbroken persona. Whether that behaviour holds at low reasoning effort, in languages other than English, against automated attacks rather than a persona prompt, and under a harness that rewrites the system prompt, is unmeasured by anyone.

• The next checkpoint's paperwork — the October 22 report settles the registration question for weights that are already public, because the draft catalogue excludes technology already in the public domain. It does not settle the release after this one. The catalogue is a draft open for comment until November 15, its thresholds are reported rather than published in the text, and the interim registration duty said to have taken effect on October 16 has not been published in full. V4.1F-1017 itself is out, MIT-licensed and ungated; what a future DeepSeek checkpoint has to clear before it can be downloaded is a rule that does not exist yet.

The beta was a two-day question about whether a Flash-tier model could carry Pro duty. The release is DeepSeek answering it with a rate card and a routing plan — and then withdrawing the routing plan within two days when developers objected to having a production model swapped underneath them. That is a weaker statement than the announcement implied and a more honest one: the quality evidence for the move is DeepSeek's own benchmarks plus outside measurements that mostly point the same way. The newest of them, a cyber benchmark run under seven agent harnesses, points the other way on the one comparison it makes, and the company has left the choice with its customers rather than making it for them. Since the release the weights have gone from a promise to a 510 GB download, the model has gone from a vendor endpoint to something you can route to, and the serving stack has gone from a day-0 recipe to a measured one — vLLM reporting itself 1.9× faster at low concurrency three weeks in, then publishing, the day the checkpoint changed, a cost curve that prices the model at 1.3B tokens per dollar of total cost of ownership on SemiAnalysis AgentX. This week the layer under all of that opened as well: DeepSeek published DeepInfer, the engine its own service runs, with a decode curve for this checkpoint, the production figures behind that curve, and an unpublished September benchmark that puts Huawei's 950DT five to seven times behind NVIDIA per chip at the speed DeepSeek actually serves at. Five weeks in, the checkpoint itself moved once: October 17 brought new weights behind the same name and the same price, with a vendor table that is better on every row it publishes and a different harness under one of them. Two days later the outside measurement arrived, and it is the most useful number in this piece: Artificial Analysis put the 1017 weights at 45 on its index — level with GLM-5.3 for second among open-weights models, a point behind MiMo-V2.6-Pro — and 33.8 on Terminal-Bench 4.0, against the vendor's 42.3 on the same benchmark — so the harness that flatters the model in DeepSeek's own table is not the harness the number survives. If you are choosing today, the honest position is that V4.1 Flash looks like the best value in the DeepSeek line, has four independent instruments to its name — none of them a reasoning score, one of them now run against the checkpoint DeepSeek has served since October 17 — a fifth evaluation under construction, no production track record, a serving stack whose two strongest numbers — a speed delta and a cost curve — both come from the project that wrote the runtime, and an outside cyber result that already disagrees with itself depending on which harness you wrap around the model, a free 200,000-token tier that appeared on one platform on October 15, and a first round of hands-on reports describing a model that checks its own work further than most callers will ask it to — which is a reason to point a slice of traffic at it, not a reason to migrate on the announcement. The update DeepSeek announced on October 17 changes the weights and adds a vendor claim — faster task completion, shorter outputs — and the first outside measurement of those weights supports half of it at best: cost per index task cut by 46%, while time per task and output tokens per task barely moved. A day on, the widest-read claim about that checkpoint was an executive's: 77 on DeepSWE v1.1 at 83× lower cost, from a Databricks staffer who also says the checkpoint is coming to that platform. Half of it survives contact with the independent boards, which now put this model's name above Opus 5 on that benchmark. The other half does not yet: the 77 is DeepSeek's own row, unreproduced, and the 83× is arithmetic with no published per-task cost behind it. Databricks' own documentation confirms the model is on its platform and that the two DeepSeek models below it are scheduled to leave; it does not confirm the checkpoint is arriving. That claim is now the second-most-interesting one about the checkpoint rather than the most. On October 19 a tracker account that has called several of these releases early declared itself “dead certain” of the 42.3 on Terminal-Bench 4.0, on the grounds that it beats GLM-5.3, and asked Artificial Analysis to run the model in a native harness. The site had already done it. On its own harness the same weights score 33.8 — below GLM-5.3's 41.9, not above it. The vendor row is real and it is DeepSeek's; the inference drawn from it does not survive being run by somebody else.

إذا كانت النسخة التي تريدها من هذا هي التي تُبقي خياراتك مفتوحة، فالثلاثة جميعًا لا يفصلها عنك سوى مفتاح واحد، وبأسعار DeepSeek نفسها مع إضافة هامش ربح بنسبة 0%. الإصدار نفسه: DeepSeek V4.1 Flash على OrcaRouter — سياق بمليون رمز، وسقف إخراج 384K، وإدخال صور، بسعر $0.15 لكل مليون رمز إدخال و$0.60 لكل مليون رمز إخراج خارج أوقات الذروة. النموذج الذي يتقاعد: DeepSeek V4 Flash على OrcaRouter. والنموذج الرائد الذي قررت DeepSeek الاستمرار في تقديمه: DeepSeek V4 Pro على OrcaRouter. وجّه شريحة من حركة المرور إلى النموذج الجديد، وأبقِ نموذجًا مُجرّبًا خلفه كخيار احتياطي للتجاوز عند الفشل، ودع التوجيه يقرر.

اختيار النموذج هو أيضًا مسألة تتعلق بما حدث من أخطاء سابقًا: فالحوادث الموثقة للذكاء الاصطناعي وإخفاقات الوكلاء لنماذج من هذه الفئة مُجمّعة في OrcaRouter AI Incident Archive، حيث يحمل كل مدخل مصادره الخاصة ودرجة خطورته ودرجة ثقته وعلامته المتنازع عليها.