Thinking Machines releases Inkling 2 Small, plans open weights next week

Thinking Machines Lab, the AI startup led by former OpenAI CTO Mira Murati, has released Inkling 2 Small, a model with 594 billion parameters that accepts image inputs and outputs only text. The company says the model is stronger at coding and easier to use inside agent loops. The model was proprietary until now, and the company says it will release the weights next week.

Artificial Analysis gave Inkling 2 Small a score of 31 on its Intelligence Index, which ties it with Gemini 3.8 Flash and puts it above average for its class. Gemini 4 Carbon and Claude Opus 5.5 top the index at 60 and 59. The model's price of $0.30 per million input tokens is above the class median of $0.24, and the reviewer called it notably slow per task and very verbose. Compared with its predecessor, it gained 41% in automation and 20% in coding.

On the AA-Omniscience v1.1 test, the hallucination rate fell from 61% to 27%, accuracy rose from 32% to 34%, and the attempt rate dropped from 77% to 53%. Artificial Analysis said that if the weights are released as planned, Inkling 2 Small would be the highest-scoring open-weight model developed outside China based on current rankings.

If the weights ship as planned, developers building agents will have a text-output open model that Artificial Analysis places among the strongest for coding and automation. Test its latency and token use in your own agent loops before adopting it, since the reviewer flagged it as slow and verbose. The lower hallucination rate may matter for tool-using agents, but the figures come from a single third-party benchmark.