Mira Murati’s Startup Releases Inkling, Largest US Open-Weight AI Model

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has unveiled Inkling, an open-weight AI model with 975 billion parameters, challenging the closed-model approach of rivals.

This article was drafted with AI assistance from multiple sources and was reviewed and approved by a human editor before publication.

Thinking Machines Lab, the artificial intelligence startup launched by former OpenAI chief technology officer Mira Murati, has released its first proprietary AI model, named Inkling, on July 15, 2026. The model is distributed under an Apache 2.0 license, making its weights freely available and positioning it as the largest open-weight model from a US company to date.

Inkling uses a mixture-of-experts (MoE) design with 975 billion total parameters, of which approximately 41 billion are active for each task. The system incorporates 256 routed experts and two shared experts, activating six experts per token generated. The model was trained on 45 trillion tokens covering text, images, audio, and video, and can reason across all four modalities, though its outputs are restricted to text. It supports a context window of up to 1 million tokens.

The training was carried out on Nvidia GB300 NVL72 systems, building on a partnership with Nvidia announced in March 2026 to deploy a gigawatt of computing capacity using Nvidia's Vera Rubin architecture. Thinking Machines claims that Inkling matches Nvidia's Nemotron 3 Ultra on the Terminal Bench 2.1 benchmark while using roughly one third the tokens. However, the company acknowledges that Inkling is not the most powerful model available today, whether open or closed.

Inkling is available on Thinking Machines' Tinker platform as of July 15, 2026, and will be accessible through third-party services including TogetherAI, Fireworks, Modal, Databricks, and Baseten. The model weights can be downloaded from Hugging Face, and it supports inference engines such as vLLM, SGLang, Miles, TokenSpeed, and Llama.cpp.

The company also previewed a smaller variant, Inkling-Small, which has 276 billion total parameters and 12 billion active parameters. For early post-training data generation, Thinking Machines used other open-weight models, such as Moonshot AI's Kimi K2.5, but says its next model will rely entirely on self-contained post-training.

Thinking Machines claims it brought Inkling to market in about nine months, compared to roughly five years for OpenAI and three years for Anthropic. A reported $50 billion fundraising round stalled by January 2026, and the company has declined to discuss funding since.