Inkling: Thinking Machines Lab's First Open-Weights Model

Thinking Machines Lab releases Inkling, a 975B-parameter open-weights multimodal model with 41B active parameters, 1M token context window, and Apache 2.0 license.

Thinking Machines Lab — founded by former OpenAI CTO Mira Murati — has released Inkling, its first open-weights model. It's a Mixture-of-Experts transformer with 975 billion total parameters (41 billion active per token), trained from scratch on 45 trillion tokens of text, images, audio, and video.

Inkling is natively multimodal, accepting text, image, and audio inputs and generating text output. It supports a context window of up to 1 million tokens and uses a 66-layer decoder-only architecture with 256 experts — each token routed to 6 experts plus 2 shared. The model is released under the Apache 2.0 license and is available for fine-tuning on Tinker, the company's customization platform.

The company describes Inkling as "not the strongest overall model available today" but positions it as a balanced, flexible foundation model designed for customization. Alongside the full model, Thinking Machines is previewing Inkling-Small, a lighter 12B-active-parameter variant trained with a similar recipe.

The release landed at the top of Hacker News, drawing over 900 points and more than 200 comments from developers eager to experiment with the weights.

Sources: Thinking Machines Lab, Inkling Model Card