Nemotron 3.5 Super VL: NVIDIAs offenes 120B-Vision-Modell ist da, und die 5x-Behauptung hat noch keine Bestenliste

N​VIDIA N​emotron 3.5 Super VL shipped on October 15, 2026 — an open 120-billion-parameter vision-language model with open weights, open datasets and training recipes, and a claim attached to it that nobody can check yet. The company's launch post calls it "the fastest open agentic model in its class — 5x faster than the next best," and attributes the number to the Artificial Analysis Intelligence Index leaderboard. As of this writing, Artificial Analysis does not list Super VL at all; its N​emotron 3.5 entries stop at N​emotron 3.5 Lightning. That does not make the claim false. It makes it vendor-reported, and it should be read that way until a third party runs the checkpoint.

The other thing worth knowing up front is a three-day gap. The weights went up on Hugging Face at 2026-10-12T15:32Z — the BF16 post-trained build, the base model and the NVFP4 build all within a minute of each other — and the announcement post did not appear until October 15 at 15:00Z, with the model card's own release-date field still reading 10/15/2026. N​VIDIA uploaded first and announced later. That is now the normal shape of an open-weights release: the repository is the primary source, and the blog post is the commentary.

Was tatsächlich ausgeliefert wurde und was nicht.

Drei Super VL Checkpoints, alle jetzt live:

• N​VIDIA-N​emotron-3.5-Super-VL-120B-A12B-BF16 — the full-precision post-trained model, the reference build and the one to fine-tune from.

• N​VIDIA-N​emotron-3.5-Super-VL-120B-A12B-NVFP4 — the quantized deployment build, produced with N​VIDIA Model Optimizer and quantization-aware distillation from the BF16 model.

• N​VIDIA-N​emotron-3.5-Super-VL-120B-A12B-Base-BF16 — the pre-post-training base, for anyone who wants to run their own post-training rather than adapt N​VIDIA's.

There is a fourth repository, N​VIDIA-N​emotron-3.5-Super-VL-120B-A12B-NVFP4-LiLiCorr, which is not a model you prompt. It is a speculative-decoding draft checkpoint: load it alongside the NVFP4 target and it proposes tokens ahead of the main model. Its own card dates it 10/15/2026 and describes it as intended for LiLiCorr deployments tuned for DGX Spark and low-concurrency data-center work.

Two things did not ship, and each matters. There is no price. N​VIDIA's announcement states the model is available as an N​VIDIA NIM microservice — a production microservice the company says is refreshed every two weeks with security patches and performance work — but publishes no per-token figure anywhere, for NIM or for the third-party platforms carrying it. And there is no independently measured speed number, for the reason above. Everything in the next two sections is N​VIDIA's own reporting.

Die Architektur in einem Absatz, weil sie die Deployment-Story erklärt.

Super VL ist ein Mamba-2-Transformer-Hybrid-Mixture-of-Experts: verschachtelte Mamba-2-, MoE- und Attention-Schichten, 512 geroutete Experten pro MoE-Schicht, plus einen C-RADIO-Vision-Encoder und einen Vision-Language-Connector, der an das Sprach-Backbone angebunden ist. Insgesamt 120B Parameter, 12B pro Token aktiv. Der Kontext reicht bis zu 1M Token, und Multi-Token-Vorhersage ist nativ integriert, was der LiLiCorr-Draft-Checkpoint ausnutzt. Eingaben sind Text, Bilder und Video — Video als MP4 bis zu zwei Minuten, als Frames abgetastet — und Dokumente können als Seitenbilder mit begleitendem Text eingespeist werden. Die Ausgabe ist ausschließlich Text.

That shape is why the deployment table is the most useful part of the documentation. N​VIDIA's BF16 build wants 4× H100 or 2× GB200. The NVFP4 build wants 1× DGX Spark, 1× H200, 2× H100, or 1× GB200. The gap between those two lines is the whole reason the quantized checkpoint exists, and it is the difference between "a 120B model you rent" and "a 120B model on the desk."

The benchmarks N​VIDIA published, and the two problems with them

Label all of this vendor-reported. N​VIDIA evaluated Super VL through NeMo Gym and published the recipes, containers, prompts and scoring settings — which is more than most labs do — but N​VIDIA trained the model, chose the baselines and ran the harnesses. Nothing here has been reproduced outside the company.

The agentic numbers are the ones the launch leans on. SWE-bench Verified 71.10, SWE-bench Multilingual 67.72, SWE-bench Pro 51.78, tau2-bench retail 69.24, airline 76.50, telecom 73.19, BrowseComp 78.91, and GraphWalks 89.71 at the over-1M-token setting. On reasoning, GPQA-Diamond 89.52 and HLE without tools 31.79, rising to 46.01 with tools — that last figure, N​VIDIA notes, comes from a separate internal harness rather than the standard path.

On vision, which is the new capability relative to the text-only N​emotron 3.5 Lightning from August: CharXiv-RQ 80.82, MathVision 78.16, V* 85.60, OCR-Reasoning 65.08, OCRBench-v2 at 68.12 English and 59.15 Chinese, MMLongBench-Doc 62.82, OSWorld 52.14, and Video-MME-v2 32.85.

Jetzt die zwei Probleme.

First, the Terminal-Bench 2.1 figure is stated twice and does not agree with itself. The model card reports 66.70. The technical report's summary table reports 60.67. Everything else in those two documents lines up, so this is not two different models — it is one number carrying two values in N​VIDIA's own paperwork, and until the recipes are re-run it is not safe to quote either. That is the sort of thing a single independent replication settles in an afternoon, which is exactly why the absence of one matters this week.

Second, the headline efficiency claim and the headline intelligence claim pull in opposite directions, and only one of them is checkable. N​VIDIA says Super VL reaches accuracy on multimodal agentic tasks "comparable with N​VIDIA's largest reasoning model at roughly one-fifth the parameter count" — N​emotron 3 Ultra 550B A55B has 55B active parameters against Super VL's 12B, a ratio of 4.6×, and the report says Super VL's 51.78 on SWE-bench Pro does surpass Ultra. Fine.

But look at the columns where Super VL is genuinely behind. MMLU-ProX 76.91, against Qwen3.5-122B-A10B at 83.2 and Qwen3.8-Flash-Next at 83.0. AA-Omniscience accuracy 26.20, which is close to the floor of that benchmark and sits below DeepSeek-V4-Flash-0731's 45.0. Video-MME-v2 32.85 against Qwen3.8-Flash-Next at 45.1. BabyVision 29.60 against 46.8. Agentic-terminal at 60.67 or 66.70 against 85.80. Super VL is not the accuracy leader in this weight class on most rows; it is a 12B-active model that N​VIDIA is selling on speed and on the breadth of what it ingests in one call, not on beating everything at everything.

Die Effizienzzahlen, die eigentlich die interessanten sind

Lass die Intelligenzvergleiche weg, und die technischen Zahlen sind konkret:

• Speculative decoding — N​VIDIA's post-training MTP boosting lifts average acceptance length from 4.287 to 4.718 tokens under greedy decoding and 3.787 to 4.612 under probabilistic draft sampling.

• Quantization — the released NVFP4 checkpoint recovers 99.62% of BF16 accuracy as a median across 36 benchmarks, so the deployment build is close to lossless by N​VIDIA's own measure.

• Durchsatz — bis zu 4.856 Ausgabe-Token/s/GPU auf einer einzelnen GB200 bei einer agentischen Proxy-Workload mit 20 Turns und gleichzeitiger Aufrechterhaltung von mindestens 50 Token/s/Nutzer; NVFP4 mit 2,43× BF16-Durchsatz unter denselben Bedingungen.

• Relativer Durchsatz — 2,34× Qwen3.5-122B-A10B und 1,90× Qwen3.8-Flash-Next bei dieser Arbeitslast.

• Serving stack — on N​VIDIA Dynamo, N​VIDIA claims up to 8× more concurrent users at 85% lower latency.

Jede einzelne davon ist eine Herstellerangabe. Diejenige, die sich für jeden am einfachsten fälschen lässt, ist die Zahl von 4.856 Tokens/s/GPU, denn sie ist eine Eigenschaft einer bestimmten GPU und einer bestimmten Arbeitslast, nicht die eines Benchmark-Laufs hinter einem Scoring-Harness.

Es bereitzustellen – und was die beiden Builds Sie an Hardware kosten

Both vLLM and SGLang are named as supported runtimes, and N​VIDIA ships working launch configurations: four-way tensor parallelism with three MTP draft tokens on 4× H100 for BF16, two-way with five draft tokens on 2× GB200. Hardware coverage is Hopper and Blackwell, Linux only, and N​VIDIA lists DGX Spark, DGX Station, RTX PRO, RTX and Jetson on the client side — the same silicon spread the model is positioned against.

Die praktische Entscheidung für ein Team, das dies am Starttag liest, ist, welchen Checkpoint es herunterladen sollte. Wenn Sie Fine-Tuning betreiben, ist der BF16-Build der einzige sinnvolle Ausgangspunkt, denn der NVFP4-Build hat bereits eine quantisierungsbewusste Destillation durchlaufen, und weiteres Training lässt sich nicht sauber damit kombinieren. Wenn Sie ihn im Serving einsetzen, ist der NVFP4-Build derjenige, der auf eine einzelne H200 oder einen DGX Spark passt, und eine mediane Wiederherstellung von 99,62 % macht ihn zur Standardwahl, es sei denn, Sie haben eine spezifische Regression gemessen. Das Basismodell herunterzuladen ergibt nur dann Sinn, wenn Sie Ihre eigene Post-Training-Pipeline schreiben wollen – das ist ein anderes Projekt.

Wofür Super VL tatsächlich gedacht ist

Read N​VIDIA's own framing and the target is narrow and specific: always-on enterprise agents that digest visual documents. The announcement's example is a web page — extract the text, read the charts, return findings — and the same claim is extended to PDFs, tables and structured documents. The argument for one model instead of a text model plus a separate vision pipeline is context continuity: the document, the data and the decision stay in the same window rather than being handed between systems.

Dieses Argument ist auf dem Datenblatt glaubwürdig, weil das Kontextfenster 1 Mio. Tokens umfasst und der Vision-Encoder gemeinsam trainiert wird, nicht nachträglich angeflanscht. Ob es den Kontakt mit echten Dokumenten bei 1 Mio. Tokens übersteht, ist genau die Frage, auf die die Zahlen von GraphWalks und MMLongBench-Doc abzielen, und beide sind Anbieterangaben.

N​VIDIA also lists a long adoption roster in the launch post — Scale AI building agents for Skydance and EY, AfterQuery for Hebbia's credit-agreement analysis, AI21 and Alice on cybersecurity, Mercor on coding agents, plus Atlassian, CrowdStrike, Dassault Systèmes, Perplexity, Salesforce and Synopsys among integrators. Those are partnerships named in a press release, which is a statement of intent rather than evidence of production traffic. Treat the roster as a signal about which verticals N​VIDIA is targeting, not as a reliability claim.

So probieren Sie es aus, ohne einen Produktionspfad darauf zu verwetten

Die ehrliche Haltung diese Woche ist, dass Super VL ein interessantes, gut dokumentiertes, vollständig unverifiziertes offenes Modell ist, und die richtige Art, ein unverifiziertes offenes Modell zu bewerten, ist nicht, ihm am ersten Tag Produktions-Traffic zuzuleiten.

Locally, the NVFP4 build on a single H200 or a DGX Spark is the cheap test: pull the checkpoint, run your own document set through it, and compare against whatever vision model you use today. That costs you GPU time and nothing else, and it is the only way to get a number that is not N​VIDIA's. N​VIDIA's own documentation gives you one free head start: the model card and the technical report both specify temperature 1.0 and top_p 0.95 for text, and temperature 0.6 with top_p 0.95 for vision — a starting point you would otherwise spend a day rediscovering.

If you would rather test it behind an API, the availability line matters: Super VL is on Hugging Face, as N​VIDIA's own NIM microservice, and through several third-party platforms. N​VIDIA has published no price for any of those paths, so any per-token figure you see this week is set by the platform, not by the vendor. On the OrcaRouter side, N​emotron 3.5 Super VL is not on our catalogue — we do not serve it today, so there is nothing here to try it through yet, and we would rather say that than imply otherwise. Everything we do serve sits behind one key across 200-plus models at provider list price with 0% markup, and where a model is available from more than one provider, automatic failover across providers gives you a way to route to a new model on a fraction of traffic and let it fall back to the incumbent when it stumbles. That is the mechanism this kind of launch wants; it is not an offer to serve this particular checkpoint.

One more piece of the release worth a look regardless of whether you want the model: N​VIDIA published its post-training datasets alongside it — spans covering agentic tool use, finance, instruction following, long context, fairness and vision-language understanding — plus the RL environments in NeMo Gym. The dataset release is often more reusable than the checkpoint, and it is the part of this launch that costs N​VIDIA the most to give away.

Was zu sehen

Vier Dinge würden diesen Artikel verändern, grob geordnet danach, wie sehr sie ins Gewicht fielen.

Ein erster unabhängiger Benchmark, von Artificial Analysis oder sonst jemandem, der entweder die Zeile für Super VL liefert, die die 5x-Behauptung impliziert, oder dies verweigert. Bis es ihn gibt, ist die zentrale Effizienzbehauptung des Launches ein Satz in einer Pressemitteilung, dem der Name eines Leaderboards angehängt wurde, und der Vorbehalt gehört in denselben Atemzug wie die Zahl.

A price. N​VIDIA has told you the hardware, the runtimes and the licence, and withheld the one figure a buyer needs to compare this against a closed model. When NIM pricing appears, the comparison becomes arithmetic instead of positioning.

The Terminal-Bench 2.1 discrepancy — 66.70 on the model card against 60.67 in the technical report. Whichever is right, one of N​VIDIA's two primary documents is wrong, and a corrected re-run would settle several other rows at the same time.

And the walk from "available now" to "actually served." Super VL's benchmark rows compare it against Qwen3.5-122B-A10B, Qwen3.8-Flash-Next, DeepSeek-V4-Flash-0731, GLM-5.3-Flash and N​emotron 3 Ultra 550B A55B — all of which are mature enough to have measured throughput, real prices and production track records. A 12B-active model that ingests video, 1M-token documents and tool calls in one pass is a genuinely useful shape. Whether it earns a routing rule is a question about latency and price stability under load, and N​VIDIA has published the first of those only on its own hardware, under its own workload, in its own units.

Benchmarks veralten und Preise bewegen sich; die Fehlerhistorie nicht: Die dokumentierten KI-Vorfälle und Agentenfehler hinter Modellen dieser Klasse sind gesammelt im OrcaRouter AI Incident Archive, wo jeder Eintrag seine eigenen Quellen, seinen Schweregrad, seinen Konfidenzgrad und eine Kennzeichnung als umstritten enthält.