Thinking Machines Releases Inkling, an Open 975B-Parameter MoE Model

Mira Murati's startup Thinking Machines Lab released its first open-weights model, Inkling, on July 16, 2026. It is a Mixture-of-Experts transformer with 975B parameters (41B active) under the Apache 2.0 license, trained on 45T tokens of text, images, audio and video.

What is Inkling

Inkling is a multimodal MoE model with 975B parameters, of which 41B are active for each request. Its developer, Thinking Machines Lab, opened the weights under the permissive Apache 2.0 license and shipped the release with an unusually short model card, far more concise than what American AI labs typically publish.

• Developer: Thinking Machines Lab, the startup of Mira Murati, former CTO of OpenAI
• Architecture: Mixture-of-Experts, 975B parameters in total, 41B active
• Data: 45T tokens of text, images, audio and video
• License: Apache 2.0, fully open weights
• A smaller version, Inkling-Small, has been promised: 276B parameters (12B active), with weights to be released after testing

Not a flagship, but a base for fine-tuning

Thinking Machines Lab openly acknowledges that Inkling is not the strongest model on the market, open or closed. Instead of racing for first place, the team positions it as a convenient base for customization and fine-tuning on its own platform, Tinker.

"Inkling is not the strongest model available today, open or closed.

What makes it a good open base for customization is a combination of qualities: multimodality, efficient 'thinking,' and availability in Tinker for fine-tuning," the Thinking Machines Lab model card says.

Documentation on the training data, however, is almost empty. The company says only that it used content from the open internet and publicly available repositories, that some data came from third parties, and that the data may include material protected by intellectual property rights.

Why this matters for the open-model market

Inkling strengthens the American open-weights ecosystem, where Chinese releases have recently set the pace. According to developer Simon Willison, who tested the model through the API, Inkling looks competitive against open models from China and joins the few viable American contenders, NVIDIA Nemotron and Gemma 4.

Multimodality holds up in practice. Willison generated an SVG image through the API, then asked the same model to describe it. Inkling picked out the image's details and took the drawn bird for a stork or a gull.

What this means

An open multimodal base from a team of former OpenAI executives lowers the barrier for anyone who wants to build and fine-tune their own models without depending on closed APIs. Thinking Machines is betting not on benchmarks but on ease of customization through Tinker.

Frequently asked questions

What is Inkling?

Inkling is the first open-weights model from Thinking Machines Lab: a multimodal Mixture-of-Experts transformer with 975B parameters (41B active), trained on 45T tokens and released on July 16, 2026.

How many parameters does Inkling have?

975B in total, of which 41B are active for each request. The promised smaller version, Inkling-Small, has 276B parameters (12B active).

Can Inkling be used for free?

Yes. The weights are open under the Apache 2.0 license, so the model can be freely downloaded, used and fine-tuned, including on the Tinker platform from Thinking Machines Lab.