Thinking Machines Lab Releases Inkling 2 Small Weights Under Apache 2.0

Thinking Machines Lab made the weights of its 594B-parameter Inkling 2 Small Mixture-of-Experts model (17B active parameters) downloadable on Hugging Face under Apache 2.0, about a week after its closed-weights launch on Tinker. The company says the delay let external safety testing and red-teaming finish first, and it reports an Artificial Analysis Intelligence Index score of 31 with strength in agentic coding, vision, and 3D design. The model also tops the open-weights ranking on DesignArena and is available through SGLang, vLLM, Unsloth, and several API providers.

Why it matters: The staged closed-then-open release is a new template for frontier open-weights launches, and the Apache 2.0 license matters to teams deciding whether they can realistically self-host a model this large.