On November 5, 2026, Thinking Machines Lab released Inkling 2 Small, a mixture-of-experts model with 594B total parameters and 17B active, supporting variable effort. It accepts images through a 292M-parameter vision encoder but not audio. At launch it is available on Tinker, the company’s platform, for inference, fine-tuning and sampling, and in the Tinker Playground.
The weights are initially closed. The company says it intends to release the full weights under Apache 2.0 “next week,” contingent on external safety testing and red-teaming, and that the model will reach additional platforms in the coming weeks.
Thinking Machines reports a score of 31 on the Artificial Analysis Intelligence Index (30.6 in its table) and leads its comparison table on HLE-Diamond (28.7%), MMMU Pro (81.6%) and ScreenSpot Pro (83.8%, or 91.5% with Python). On coding it trails DeepSeek-V4.1-Flash-1017, for example 72.1% versus 77.0% on DeepSWE v1.1 and 25.4% versus 28.5% on FrontierCode v2.0.
Why it matters: Thinking Machines is using a release-then-open sequence - hosted first, weights after outside safety testing - which is a concrete version of the “safe path to open weights” approach the lab described in July.
What it does not show: all scores are self-reported, the weights had not been released as of the announcement, and the page gives no per-token price.