Thinking Machines released Inkling-Small, an openly downloadable model that accepts text, images and audio and replies in text. It is a smaller companion to the firm's larger Inkling model, using 12 billion of its 276 billion parameters for any given piece of text, and it can be run locally through common serving libraries or reached through hosted providers. The publisher notes it still makes things up at times and advises added filtering and human review before use in sensitive settings.
What changed
Thinking Machines' Inkling family previously offered only the larger model, with 41B active of 975B total parameters.
What it unlocks
Running a smaller open-weight model locally that takes text, images and audio and answers in text, and fine-tuning it for a specific product.
- 276B total parameters, 12B active
- SWE-Bench Verified 80.2%
- GPQA Diamond 89.5%
- AIME 2026 95.5%
- 32,829 downloads last month
What you need to act on it
- hardware or an inference provider able to host a 276B-parameter model
- acceptance of the model acceptable use policy
Sources