Thinking Machines Lab, the company founded by former OpenAI CTO Mira Murati, has released Inkling 2 Small, a Mixture-of-Experts model with 594 billion total parameters, 17 billion active, image inputs, up to 1M tokens of context, and an adjustable reasoning effort. It scores 31 on the Artificial Analysis Intelligence Index, well above the first Inkling generation, with the largest gains in agentic coding, vision and knowledge work, and it uses vision in the loop to check and fix the interfaces it generates.

On the November 5 DesignArena Website leaderboard the model holds 1,340 points, third overall behind only the closed Muse Wonder (1,366) and Muse Spark 1.3 (1,351), and it is set to be the top open-weights entry once its weights ship. For now it runs on Tinker for inference and fine-tuning with closed weights; Thinking Machines plans to publish the full checkpoints under Apache 2.0 next week after external safety testing and red-teaming. Despite the name, 17 billion active parameters do not shrink the 594 billion weights enough to run on an ordinary desktop.

Related: Thinking Machines Lab Releases Inkling, Their First Open-Weights Multimodal Model, Meta Releases Muse Wonder Agentic Models

Introducing Inkling 2 Small