Thinking Machines launches Inkling 2 Small on Tinker, with open weights promised next week

Thinking Machines introduced Inkling 2 Small on November 5, the first model of its second generation. It is live on Thinking Machines' Tinker platform today, in Tinker Playground and for fine-tuning and sampling, with closed weights. The company says the full weights will follow next week under an Apache 2.0 license, after external safety testing and red-teaming.

THE MODEL

A mixture-of-experts model with 594B total parameters and 17B active, with image input and variable effort. The first generation was Inkling (975B, July 15) and Inkling-Small (276B total, 12B active, July 30).

THINKING MACHINES' OWN FIGURES

- A score of 31 on the Artificial Analysis Intelligence Index, which the company says far surpasses its previous models.

- FrontierCode v2.0: 25.4%. DeepSWE v1.1: 72.1%. SWEBench Pro v2 (Hard): 86.3%.

- In the same table, DeepSeek V4.1 Flash 1017 scores higher on FrontierCode (28.5%) and DeepSWE (77.0%), and Kimi K3 higher on SWEBench Pro v2 (Hard) (88.2%).

THE STEALTH NAME

Thinking Machines says it tested early checkpoints on OpenCode under the name Fledge-Alpha, one of the anonymous models that had been circulating there.

WHAT WE DO NOT KNOW

Tinker pricing for the new model, its context length, and whether the weights land on the date promised. The Artificial Analysis score is Thinking Machines' statement; we have not read it on Artificial Analysis's own site yet.