Inkling: Our open-weights model
Simon Willison · Simon Willison · 2026-07-16
Thinking Machines Lab, founded by Mira Murati, releases Inkling—a 975B-parameter Apache-2.0 licensed open-weights multimodal Mixture-of-Experts model trained on 45 trillion tokens—positioning it as a customizable fine-tuning base rather than a frontier model competitor.
Appears in
Extraction
Topics: open-weights-modelsmixture-of-expertsmultimodal-modelsllm-releasethinking-machines-lab
Claims
- Inkling is a 975B total parameter (41B active) MoE transformer trained on 45 trillion tokens of text, images, audio, and video, released under the permissive Apache-2.0 license.
- Thinking Machines explicitly positions Inkling as a fine-tuning base rather than the strongest available model, tying its value proposition to their Tinker training and customization platform.
- A smaller companion model, Inkling-Small (276B total, 12B active), is still in testing with weights to be released once evaluation is complete.
- The model card and training data documentation are notably sparse by US AI lab standards, offering only generic descriptions of 'publicly available content' without meaningful data sourcing detail.
Key quotes
There's a lot to like about this release. It's Apache-2.0 licensed, and looks competitive with the open weight models coming out of China - it's good to see the US open weights ecosystem gain a new viable contender to join NVIDIA Nemotron and Gemma 4.
Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning.