一张生成的标题卡,标题为“Inkling 2 Mini vs Qwen 3.8”,副标题为“真正参与竞争的是 Flash 这一档”,三个胶囊标签分别写着“Mini:Index 20.1”“Flash:Index 29”和“每任务成本相同”,下方是两张卡片,分别对比“20GB 文件,可离线运行”与“托管端点,1M 上下文”,并附注仅在需要时升级。OrcaRouter 标志位于右下角。
Guides & Insights

Inkling 2 Mini 对比 Qwen 3.8:被更便宜的档位击败

作者

jinhao.song

发布日期

最新模型 · 20查看全部模型 →
基准测试:Artificial Analysis · 每日更新
返回全部文章

大家所进行的那项比较是错的,而且这项比较还抬高了新模型。Inkling 2 Mini 由 Thinking Machines Lab 于 2026 年 11 月 20 日发布,在 Artificial Analysis Intelligence Index v5.0.1 上得分 20.1——与 Qwen3.8-27B 的 20 分持平,但每项指数任务成本为 0.59 美元,而后者为 3.53 美元。这正是 Thinking Machines 放进自家基准表中的正面对比,而在成本上,两者差距并不接近。但 Qwen 3.8 并不是一个模型。它是一个包含五个模型的世代,而真正能与一个 20GB 可下载检查点竞争的档位是 Qwen3.8-Flash;Artificial Analysis 给它的评分是 29,每项指数任务成本为 0.57 美元——领先九分,而每完成一个答案的成本相同。Mini 赢下了发布帖中印出的那项比较,却输掉了买家真正会进行的那项比较。

Sourcing, since it drives the verdict: the index scores, cost per task, latency and context figures are Artificial Analysis measurements read on 21 November 2026, and every OrcaRouter price is quoted from our own catalogue. The percentage tables are Thinking Machines' launch harness, vendor-run, flagged where they appear; A​libaba's own claims for the Q​wen tiers are quoted as A​libaba's and have not been separately audited here.

首先,你指的是哪个 Qwen 3.8?

把家族名当作对手,正是让大多数此类对比失去意义的错误。Q​wen 3.8 涵盖每百万 token 收费 2.00/6.00 美元的托管旗舰、采用自定义许可证的开放 2.4 万亿参数检查点、Apache 2.0 下的 27B 稠密模型、价格为 0.15/0.47 美元的 Flash 层级,以及 Flash-Next 预览版——其开放权重被置于托管 Flash 产品之后。各成员在 token 价格上相差十倍以上,在同一指数上相差九个百分点。单一的“对比 Q​wen 3.8”结论,只是对一个模糊对象下的结论。

• Qwen3.8-27B — 27B 稠密模型,Apache 2.0,支持文本、图像和视频输入,256K 上下文,在 xhigh 努力程度下 Index 为 20,标价 $0.50/$3.00,每个 Index 任务 $3.53,每任务 34.5 分钟。

• Qwen3.8-Flash——托管生产层级,价格为 $0.15/$0.47,所服务的权重与 Flash-Next 预览版相同;路由端点宣称提供 1M token 窗口和 131K 最大输出,而 Artificial Analysis 记录开放 Flash-Next 权重的窗口为 256K。其测得该模型的 Index 为 29,总参数 180B、激活参数 6B,每项索引任务 $0.57,耗时 33.3 分钟。

• Qwen3.8-Max —— 闭源旗舰模型,指数 34,$2.00/$6.00,每个指数任务 $10.02。在成本上不是 Mini 的竞争对手,也不是应当拿来对比的模型。

• Inkling 2 Mini — 总参数 26.4B,激活参数 3.4B,Apache 2.0,卡上 1M 上下文,Index 20.1,Tinker 上 $0.15/$0.50,每个索引任务 $0.59,9.0 分钟。

所以有用的比较有两部分。面对发布文章选中的 27B 稠密模型,Mini 在同等分数下是真正的成本优势。面对 Flash 档位——成本相同,且在指数上远超它——Mini 的论据必须关乎价格和分数之外的东西,而它确实是:文件。

Screenshot of the Artificial Analysis model page for Qwen3.8-Flash-Next, showing it above average in intelligence and well priced for its size but notably slow per task and very verbose, a score of 29 on the Intelligence Index against a median of 23, 156M tokens generated during the index run against a 104M median, pricing of $0.15 per 1M input and $0.47 per 1M output tokens, $0.57 per index task, 33.3 minutes per task, an 85.77-second time to first token, 74 output tokens per second, and a listed 256K-token context window.

27B 平局,以及为什么它是真实的

两个模型由同一评测器在同一索引的同一修订版本上测量,因此 20.1 对 20 是真实比较,而不是评测框架造成的巧合。这也是在仪器分辨率极限处的一次平局。该索引由九项评估复合而成,其中几项基于留出集,而彼此相差一分以内的模型无法由它可靠地排序。这次平局所确立的是:Mini 达到了与 27B 稠密模型相同的门槛,同时只消耗 34 亿活跃参数,而不是 270 亿。

成本这一列才是关键所在。每项索引任务为 $0.59,而对比项为 $3.53,Mini 完成同一评估的成本大约只有六分之一,而且用时九分钟,对比 34 分半。输出速度是每秒 280 个 token,对比 53 个。首个回答 token 的时间为 9.72 秒,对比 144 秒。在一个运行三十轮的智能体循环中,这些数字会累积成用户能切实感受到的差异。

None of that is free, and the compensating cost is verbosity. Mini emitted 161M output tokens to complete the index run, against 128M for the Q​wen model and a peer median of 85.7M. Paying $0.50 per million output for 1.9 times the typical output is still cheap; budgeting a fixed number of tokens per task for a model that talks this much is not. A team that meters by tokens rather than by tasks will find the six-times advantage shrinks toward three.

Where Mini genuinely beats every Q​wen tier

供应商的编程表是这次发布中让模型显得有意思的部分,也是没人检查过的部分。下面的每一行都是 Thinking Machines 自己的测试框架在其自己的检查点上的结果,而且其中至少有一行是用公司的智能体脚手架而不是中立的脚手架运行的。

• DeepSWE v1.1 — Inkling 2 Mini 48.2% 对比 Qwen3.8-27B 42.2%。

• SWE-bench Pro v2,困难 — 77.9% 对比 73.8%。

• SWE-bench Multilingual — 73.7% 对比 70.0%。

• SWE-bench Verified — 79.2% 对比 77.8%。

• Vibe Code Bench v1.1 — 77.9% 对比 64.9%。

• Terminal Bench v2.1 — 71.6% 对 73.0%,落后;在 Terminal Bench v4.0 上两者均打印 5.6%。

• BrowseComp — 79.0% 对比 61.6%;DeepSearchQA 85.3% 对比 79.7%;MCP Atlas 82.7% 对比 77.3%。

• MMMU Pro — 74.9% vs 76.3%,又一次以微弱差距落败;OmniDocBench v1.6 92.8% vs 91.5%。

Mini 胜出的那些行都是智能体类任务——仓库级补丁、长程搜索、工具使用。它落败的那些行则是精确召回类任务。这正是架构本身的体现:循环换来的是每个 token 的串行计算,而串行计算正是长智能体轨迹所需要的。它换不来存储的事实,而知识密集型基准测试会注意到这一点。

Mini 与家族之间的事实性差距比上述任何一行都更刺眼。在 AA-Omniscience 上,厂商表格中 Mini 的得分为 −8.6,非幻觉率为 63.4%,而同一表格同一列中 Qwen3.8-27B 为 −6.4、非幻觉率 71.4%。负分意味着模型的表现比拒绝作答还差。我没有找到 Artificial Analysis 单独发布 Mini 的 Omniscience 指标,因此这一行完全依赖厂商测试框架;但对任何在没有检索层的情况下部署的人来说,它仍是这项比较中最重要的数字。

Flash 终结争论之处

现在来看真正重要的这一档。Qwen3.8-Flash 已托管,定价为输入 $0.15、输出 $0.47——与 Tinker 采样 Mini 的输入价格相同,输出则便宜三美分。Artificial Analysis 给其背后的 Flash-Next 权重打分为 29。每个索引任务成本为 $0.57,而 Mini 为 $0.59。多九个索引点,还便宜两美分。

That is not a close call on capability or on price, and it reframes what Mini is for. Mini's advantages over Flash are not on the scoreboard at all: it is a 20GB file that runs with no network, on a MacBook or a DGX Spark, with no per-token meter and no provider in the loop. Flash's advantages are nine index points, a hosted endpoint, and a 1M-token window on the routed API with nothing to download. If your workload has a network connection, the Q​wen side wins on every measured axis. If it does not, nothing on the Q​wen side is available to you at all.

速度对比也更有利于 Mini,而且优势比分数差距所暗示的还要大。Mini 以每秒 280 个输出 token 的速度运行,首个 token 耗时 9.72 秒;Flash-Next 测得每秒 74 个 token,首个 token 耗时 85.77 秒,每个索引任务耗时 33.3 分钟。对于交互式智能体而言,这就是一个能及时响应的工具与一个需要你等待的工具之间的差别——而这正是有效参数量为 5.4B 的本地模型相对于 180B 托管模型具有结构性优势的唯一之处。

实际安排

Both Q​wen tiers are on OrcaRouter, which makes the escalation path unusually cheap to build. Qwen3.8-27B routes at $0.33 per million input and $2.40 per million output — below A​libaba's $0.50/$3.00 list, because the provider's price is passed through rather than marked up. Qwen3.8-Flash routes at $0.15/$0.47 on a 1M-token window. One key reaches both, with automatic failover and the routing DSL to send a request to the cheap tier first and escalate only when it fails a check.

That is the shape this comparison actually wants. Mini, if you adopt it, sits on your own hardware doing the agentic work it is good at and paying nothing per token. The Q​wen tiers sit behind a single endpoint for the requests Mini is bad at — the ones that need a fact recalled correctly, a long document read, or a hard scientific inference — and the failover means a provider outage degrades the answer instead of breaking it. We do not host Inkling 2 Mini and this piece will not pretend otherwise; what the router removes is the second contract and the second integration for the half of the workload that goes to Qwen.

Screenshot of OrcaRouter's model page for Qwen3.8 Flash, showing the model id qwen/qwen3.8-flash, vision, tools, JSON and reasoning tags, a dated description of it as Alibaba's multimodal reasoning model for coding, agentic workflows and long-video analysis, an OpenAI-compatible /v1/chat/completions endpoint at api.orcarouter.ai, an input price of $0.15 and output price of $0.47 per 1M tokens, a 1M-token context window with a 131K maximum output, a p50 time to first token of 10.00 seconds, and an option to add it to a comparison.

什么会改变判决?

两件事,而且观察起来成本都很低。第一件是在 Mini 的公开权重上做一次独立编码运行。如果 DeepSWE 上的 48.2% 和 SWE-bench Pro 上的 77.9% 能在中立测试框架下保持,那么 Mini 在智能体方面的论据就会比 20.1 的指数分数所暗示的强得多,因为该指数恰恰低估了这款模型所擅长的长周期工作。如果不能保持,那发布表格就是一张厂商表格,这个模型本质上就是 20 分。

The second is whether Qwen3.8-Flash's nine-point advantage holds under effort controls. Its index figure is a single configuration, and Q​wen models move several points between effort settings — the 27B model scores 20 at xhigh and materially lower when thinking is dialled back. A Flash tier that loses nine points at low effort is a different competitor from one that keeps them, and nobody has published that curve for this comparison.

Until one of those lands, the honest summary is a split verdict. On measured capability per dollar in a hosted setting, Q​wen 3.8 wins twice — the 27B model ties Mini on score and loses on cost, the Flash tier beats it on both. On the things a hosted model cannot do at all — run offline, run on a laptop, run with no meter — Inkling 2 Mini has no Q​wen answer, and the 20GB file is the entire argument. Buy the download if you need it to run where there is no network. Buy the API if you do not.

A generated two-column scoreboard titled 'Inkling 2 Mini vs Qwen3.8-Flash — the scoreboard'. The Inkling 2 Mini column reads Intelligence Index 20.1, active parameters 3.4B, cost per index task $0.59, output speed 280 tokens per second, first token 9.72 seconds, and runs offline: yes, 20GB. The Qwen3.8-Flash column reads Intelligence Index 29, active parameters 6B, cost per index task $0.57, output speed 74 tokens per second, first token 85.77 seconds, and runs offline: no, hosted only. A footer states that both columns are per Artificial Analysis and coding rows are vendor-reported.