Thinking Machines Lab releases “Inkling-Small,” which delivers performance comparable to Inkling at nearly one-quarter the size

Thinking Machines Lab has released Inkling-Small, an open-weight Mixture-of-Experts model.

Inkling-Small has 276 billion total parameters and 12 billion active parameters, far smaller than the earlier 975 billion-parameter Inkling model.

The company says it matches or nears Inkling's performance on many benchmarks while requiring much less compute.

It supports native audio and image reasoning, variable thinking depth, and up to a 1 million token context window, and is available via Tinker and Hugging Face.