Thinking Machines' Inkling 2 Small Could Be the West's Best Open Model — but You Can't Download It Yet

Thinking Machines Lab is making a bet that the best Western open model can come from a startup, not a giant. The catch: as of Friday, it isn't open yet.

The lab, founded by former OpenAI CTO Mira Murati, introduced Inkling 2 Small on Thursday as the first model of its second generation, according to ZDNET. It's a Mixture-of-Experts model with 594 billion total parameters and 17 billion active, accepts image inputs and offers variable effort, per Thinking Machines' announcement. It's live now on the company's Tinker platform for inference and fine-tuning, with closed weights for the moment. The company says it intends to release the full weights under an Apache 2.0 license next week, after external safety testing and red-teaming.

The independent numbers are the draw. Inkling 2 Small scored 31 on the Artificial Analysis Intelligence Index, tied with Gemini 3.8 Flash, according to ZDNET. For scale, Gemini 4 Carbon and Claude Opus 5.5 top that index at 60 and 59. Artificial Analysis said that if the weights ship as planned, it would be "the highest-scoring open weights model developed outside China based on current rankings." The reviewer also reported a 41% jump in automation capabilities and 20% in coding over the previous generation, and a drop in hallucination rate on its AA-Omniscience test from 61% to 27%.

It isn't a clean sweep. Artificial Analysis described the model as "notably slow per task and very verbose," and called its $0.30 per million input tokens "somewhat expensive" against a $0.24 class median, per ZDNET.

The context matters more than the score. Chinese labs such as DeepSeek and Moonshot have dominated open-weight leaderboards since early 2025. ZDNET notes the release follows French lab Mistral's Le Chonk, which was released through the same staggered process of weeks of safety testing before the weights go out. Western open models are finally trading blows with Chinese ones.

What to watch: whether the weights actually land next week under Apache 2.0, and whether Artificial Analysis's ranking holds once developers can run the model themselves.

— via ZDNET