AI tools ·
Thinking Machines just gave away its 594-billion-parameter model for free
A week after launching it behind its own platform, Thinking Machines Lab has handed over its second-generation model. The weights for Inkling 2 Small were published on Hugging Face on November 12 under the Apache 2.0 license, which means anyone can download the model, run it on their own hardware, fine-tune it, and build commercial products on it without asking permission or paying a license fee.
The model is a mixture-of-experts design: 594 billion parameters in total, with only 17 billion active for each token it processes, so it behaves like a much smaller model when it runs. It takes text and images in and produces text, with a context window of up to one million tokens, and the lab reports a score of 31 on the Artificial Analysis Intelligence Index — a large jump over its first-generation models. The weights ship in two compressed formats, NVFP4 and MXFP8, and day-one support covers vLLM, SGLang, Unsloth, and TokenSpeed, with hosted APIs at NVIDIA, Fireworks, Together, Baseten, Nebius, Modal, and Cloudflare.
The release keeps a promise the lab made at launch on November 5, when it said the weights would follow after outside safety testing and red-teaming. One gap remains: the lab has not published the results of that safety testing. That does not mean the testing failed, but for a model this large, released openly, the results are the evidence that the week's delay bought something. Until they appear, the release is a promise kept on timing and an open question on transparency.
Why it matters
Open weights change who can use a model. A frontier-scale model that anyone can download and fine-tune is a real step for builders who cannot ship their data to a hosted API — and the Apache 2.0 license means small teams can put it in a product, not just a demo.
Sources
- Thinking Machines Lab — Introducing Inkling 2 Small (Nov 12 weights update)
- AI Agent Update — Thinking Machines publishes Inkling 2 Small weights a week after launch
Spot an error? We correct quickly and note it. Contact us via the contact page.