Models
Thinking Machines releases Inkling 2 Small, a 594-billion-parameter model with 17 billion active; its weights are promised for next week
Thinking Machines' Inkling 2 Small: 594B parameters, 17B active. Weights are closed today, Apache 2.0 promised next week. 'Best open outside China' unproven.
On November 5, 2026, Thinking Machines Lab, the AI lab led by Mira Murati, released Inkling 2 Small, a mixture-of-experts model with 594 billion parameters in total and 17 billion active per token, which accepts text and images. At release it is available only through the lab's Tinker platform, with closed weights; the lab says it intends to publish the full weights under an Apache 2.0 license next week, after safety testing. That it would be the best open model outside China is not a claim the lab makes, and it cannot be checked until the weights are out. Separately, press reports say the lab is in talks to raise $5 billion or more at a valuation of at least $40 billion; nothing is closed.
The essentials
- Model: 594 billion parameters in total, 17 billion active (mixture of experts); text and image input, variable reasoning effort and a 292-million-parameter vision encoder, per the lab.
- Access today: Tinker (inference and fine-tuning) and the Tinker Playground; weights are closed at release.
- Weights: the announcement promises Apache 2.0 weights next week, pending safety evaluations and red-teaming. The lab's Inkling family page says only that they will come in the coming weeks: two wordings, no firm date.
- Benchmarks (the lab's tables): 31 on the Artificial Analysis Intelligence Index (30.6 in the lab's table), against 13 for Inkling Small; DeepSeek-V4.1-Flash-1017 is ahead at 32.6.
- Funding: talks reported at $5 billion to $6 billion with a pre-money valuation of at least $40 billion; not closed.
Is Inkling 2 Small an open-weight model today?
No. At release the weights are closed and the model is reached through Tinker. The lab's announcement says it intends to release the full weights under Apache 2.0 next week, once its own and its partners' safety evaluations and red-teaming are complete. Until the files can be downloaded, "open" describes an intention. The earlier models, Inkling (975 billion parameters, 41 billion active, July 15) and Inkling-Small (276 billion, 12 billion active), were published with open weights. The lab also reports that this model's chemical, biological, radiological and nuclear scores sit below the open-weight frontier, that its offensive cyber results are a large step up from Inkling Small but remain behind that frontier, and that its alignment still trails closed models such as GPT-6.1 Sol.
Is it the best open model outside China?
The announcement does not say so. The nearest claim is that, once its weights qualify, it is set to be the top open-weights model on DesignArena. In the lab's own tables the picture is mixed: DeepSeek-V4.1-Flash-1017 leads several, including the Artificial Analysis index (32.6 against 30.6) and DeepSWE, and Kimi K3 leads SWE-Bench Pro Hard, while Inkling 2 Small ranks first among open-weight models on AA-AnalystAgent and Dayjob:Finance and second on the Vals Finance Index and DiligenceBench. Superpower Daily also reports 25.4% on FrontierCode v2.0 against 28.5% for DeepSeek. DeepSeek and Kimi come from Chinese labs, so "best outside China" would be a narrower ranking among non-Chinese open models. We could not verify such a ranking, and it needs the weights to be public.
What do we know about the funding round?
The Information reported, as relayed by The Next Web and others, that the lab is in talks to raise $5 billion to $6 billion at a pre-money valuation of at least $40 billion, with Nvidia reportedly in for about $2.5 billion. One summary dates these talks to September and says the round had not closed, and Forkast notes that an earlier attempt at $50 billion to $60 billion collapsed. We found no announcement of a closing. So "$5 billion at $40 billion" is the low end of a reported range, and the valuation is pre-money, not post-money.
What is not established
- The date and conditions of the weight release: "next week" is an intention, tied to the outcome of safety testing.
- Independent confirmation of the benchmark figures, which come from the lab or from outlets relaying it.
- The hardware needed to run the model yourself, which the coverage we read does not state.
- Whether the funding round closes, on what terms and with what share from Nvidia.
- Whether it ranks first among non-Chinese open models.
Summary table
| Point | What is stated as of November 6, 2026 | Source |
|---|---|---|
| Size | 594B total, 17B active, mixture of experts, text and image input | Thinking Machines Lab |
| Access | Tinker only; weights closed at release | Thinking Machines Lab, Superpower Daily |
| Weights | Apache 2.0 promised next week (family page: coming weeks) | Thinking Machines Lab |
| Artificial Analysis index | 31 (30.6 in the lab's table); Inkling Small 13; DeepSeek-V4.1-Flash-1017 32.6 | Thinking Machines Lab, MarkTechPost |
| Funding | Talks at $5-6 billion, pre-money valuation of at least $40 billion; not closed | The Next Web, Let's Data Science |
| Best open model outside China | Not claimed by the lab; unverified | Our reading of the lab's announcement |
FAQ
Can I download Inkling 2 Small now?
No. You can try it in the Tinker Playground or use Tinker for inference and fine-tuning; downloadable weights are promised for next week.
Is it the best open model outside China?
Not established. The lab does not claim it, DeepSeek leads several of its own tables, and the weights are not yet public.
Is the $40 billion valuation confirmed?
No. It is a reported floor for a pre-money valuation in ongoing talks.
Method and limits
Sources checked on November 6, 2026. We read the lab's announcement through an automated extract that covered most of the page, not every table. Benchmark numbers are the lab's; third-party coverage relays them. Funding details come from reports by The Information relayed by other outlets; we did not read the original. This is not investment advice.
Sources
- Thinking Machines Lab: Introducing Inkling 2 Small (verified 2026-11-06)
- Thinking Machines Lab: The Inkling family (verified 2026-11-06)
- MarkTechPost: Thinking Machines Releases Inkling 2 Small, a 594B MoE With 17B Active (verified 2026-11-06)
- The Tribune (ANI): Thinking Machines unveils Inkling 2 Small model with 594 billion parameters (verified 2026-11-06)
- Superpower Daily: Thinking Machines releases Inkling 2 Small through Tinker, holding back downloadable weights (verified 2026-11-06)
- The Next Web: Thinking Machines is in talks to raise $6bn, The Information reports (verified 2026-11-06)
- Let's Data Science: Thinking Machines Lab Pursues $5-6 Billion Funding Round (verified 2026-11-06)
- Forkast: Report on Thinking Machines Lab's funding talks at a valuation of at least $40 billion (verified 2026-11-06)
Spotted an error in this article? Report it · Our corrections policy
