Mira Murati's Thinking Machines Lab has released Inkling, its first open-weights multimodal model. The model is a Mixture-of-Experts transformer with 975B total parameters and 41B active parameters, trained on 45 trillion tokens of text, images, audio, and video.
- Licensed under Apache-2.0, Inkling is positioned as a strong base model for fine-tuning via the lab's Tinker platform rather than a frontier inference model.
- A smaller variant, Inkling-Small (276B total, 12B active), is currently in testing with weights to be released upon completion.
- The training data documentation notes the use of public domain content and publicly available data from the open internet and third parties.
The release aims to provide a viable US-based contender for the open-weights ecosystem, offering multimodal capabilities and efficient thinking for customization.