Thinking Machines' Inkling 2 Small model becomes the leading open-weight AI in the West

Summary

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has unveiled Inkling 2 Small (594 billion parameters, visual input/text output). The model scored 31 on the Artificial Analysis Intelligence Index. With weights set to be released next week, the model currently holds the top position among open-weight models developed outside of China.

Why it matters

Chinese labs like DeepSeek and Moonshot have dominated the open-weight model space for over a year. This dominance is forcing Western companies to reconsider expensive closed-model contracts. A reliable Western open-weight competitor shifts geopolitical and commercial balances. It offers companies a non-Chinese inference option that they can host on their own infrastructure and control the costs of. Furthermore, it demonstrates that the 'open-weight' category has evolved from a second-tier alternative into a genuine competitive tier.

Highlights

• The 594-billion-parameter model scored 31 on the AA Intelligence Index. This score is on par with Gemini 3.8 Flash; however, it trails behind Gemini 4 Carbon (60) and Claude Opus 5.5 (59).
• The hallucination rate was reduced from 61% to 27%, and automation capabilities increased by 41% compared to the previous generation. However, the model is described as 'particularly slow and verbose per task.'
• The model's price is $0.30 per million input tokens. This figure is above the category average of $0.24. The weights are not yet public. Therefore, its 'open-weight' status remains a promise rather than a reality.

Critical take

Labeling Inkling 2 Small as the 'Western open-weight leader' is generous. The model's index score of 31 is barely half that of the top-tier closed models. Moreover, Artificial Analysis notes that the model is slow, verbose, and priced above average. The real test will be whether the weights are actually released next week and whether independent benchmarks verify the 41% automation improvement outside of the lab's own evaluation cycle.