Thinking Machines Lab launches Inkling, a 975B-parameter open model

What happened: Thinking Machines Lab, founded by former OpenAI CTO Mira Murati after a $2 billion seed round, has released its first open-weights model, Inkling. Weights are on Hugging Face.

The numbers: Inkling is a mixture-of-experts transformer with 975 billion total parameters, 41 billion active per token, 256 routed experts per MoE layer with 6 active at a time. It handles up to 1 million tokens of context and was trained on 45 trillion tokens of text, images, audio and video.

The details: A smaller variant, Inkling-Small, runs 12 billion active parameters (276 billion total); its full weights will follow once testing finishes. On the lab's own benchmarks, Inkling scored 63.8% on Terminal Bench 2.1, 97.1% on AIME 2026, 87.2% on GPQA Diamond, 77.6% on SWE-bench Verified and 78.0% on the FORTRESS adversarial safety test. The lab says it hits those reasoning scores at roughly a third of competitors' token cost.

Who's affected: Inkling is available through Thinking Machines' Tinker fine-tuning platform, currently at a 50% discount, and via APIs from Together AI, Fireworks, Modal, Databricks and Baseten.

Why it matters: Thinking Machines isn't claiming the top spot. It's pitching Inkling as raw material for companies to customise.

In their words: "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 lab said in its launch post.

The rivals: The release lands the same week as Moonshot's 2.8-trillion-parameter Kimi K3, as open-weights models from both the US and China keep narrowing the gap with closed frontier systems.

What's next: Watch for Inkling-Small's full weights once Thinking Machines finishes testing it.