Thinking Machines Lab releases open-weights model Inkling

Thinking Machines Lab released Inkling, its first open-weights model, on 15 July. The lab said the model's weights are available on Hugging Face.

The company said Inkling is a mixture-of-experts transformer with 975 billion total parameters and 41 billion active per token, drawing on 256 routed experts per layer with 6 active at a time.

The model supports a context window of up to 1 million tokens and was trained on 45 trillion tokens of text, images, audio and video, the lab said.

Thinking Machines Lab also said it is preparing a smaller variant, Inkling-Small, with 12 billion active parameters out of 276 billion total, whose full weights will follow once testing finishes.

On benchmarks reported by the lab, Inkling scored 63.8% on Terminal Bench 2.1, 97.1% on AIME 2026, 87.2% on GPQA Diamond and 77.6% on SWE-bench Verified, and 78.0% on the FORTRESS adversarial safety test. The lab said its reasoning scores come at about one-third of the token cost of competitors.

The model is accessible through the lab's Tinker fine-tuning platform, currently offered at a 50% limited-time discount, and through APIs from Together AI, Fireworks, Modal, Databricks and Baseten.

Thinking Machines Lab was founded by Mira Murati, former chief technology officer of OpenAI, after a $2 billion seed round. The lab said Inkling is meant as a customisable base model rather than a top performer: "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."

The release came the same week as Moonshot's 2.8 trillion-parameter Kimi K3, as open-weights models from the US and China push to close the gap with closed frontier systems. Thinking Machines Lab has not said when Inkling-Small's full weights will be released. Full details are in the lab's launch post.