Thinking Machines Lab releases Inkling, its first open-weights multimodal model

What's the deal? Thinking Machines Lab released Inkling on July 15, 2026 — its first model, trained from scratch with the full weights made publicly available.

Inkling is a Mixture-of-Experts transformer with 975B total parameters and 41B active, and a context window of up to 1M tokens. It was pretrained on 45T tokens of text, images, audio and video. A lighter Inkling-Small preview, with 12B active parameters, ships alongside it.

Why now? Inkling reasons natively across text, images and audio, and lets developers dial thinking effort up or down to balance cost against performance. The lab positions it not as the strongest model available, but as a broad, balanced base built for customisation.

It is available for fine-tuning today on Tinker, the lab's model-customisation platform, with an Inkling Playground added to the Tinker console for hands-on testing.

The reaction: The launch drew attention as a marker of momentum for open models in the American ecosystem. Sriram Krishnan wrote that open-source models and harnesses are "having a moment", pointing to Inkling as evidence that developers can now catch up to near state-of-the-art performance with a clear training lineage.

The signal: Thinking Machines Lab was founded in 2026 by former OpenAI chief technology officer Mira Murati, and is reported to be in talks for a raise that could value it at $50B. Inkling is its first shipped product — an open-weights release that stakes a claim in multimodal reasoning against both open and closed rivals, and the first in a planned family of models.

Sources:
Thinking Machines Lab
Thinking Machines Lab, X post
Sriram Krishnan, X post