Thinking Machines Lab has launched Inkling, its first fully released open-weights foundation model family. The release positions the model as a customizable multimodal base rather than a benchmark-maxed flagship.
- Inkling is a Mixture-of-Experts transformer with 975B total parameters and 41B active parameters.
- It supports a context window of up to 1M tokens and was pretrained on 45 trillion tokens of text, images, audio, and video.
- The model reasons natively over text, images, and audio while balancing cost with performance through efficient thinking effort.
- A lighter-weight variant, Inkling-Small, is also available with 12B active parameters for lower cost and latency.
- The release includes immediate support on the Tinker platform and broad day-0 ecosystem compatibility with tools like vLLM, SGLang, and Hugging Face.
The company emphasizes that this is a day-1 foundation intended for practical use and customization, serving as the basis for future iterations rather than their final frontier push.