Mira Murati Launches Open Source AI Model Inkling After Leaving OpenAI

Mira Murati has officially released Inkling, a multimodal AI model developed by her newly founded company, Thinking Machines Lab, on July 15, 2026. The model boasts 975 billion parameters and is now fully available for free download on Hugging Face under an Apache 2.0 license. With a reported performance score of 74.1% on MCP Atlas, Inkling is positioned as a leading Western open-weight model, outperforming Nvidia's Nemotron 3 Ultra in some tasks.

Murati's journey to this point includes a significant transition from her role at OpenAI, where she served as Chief Technology Officer and briefly as interim CEO. After her departure in September 2024, she founded Thinking Machines Lab in February 2025. Shortly after, the company raised $2 billion at a $12 billion valuation, which marks one of the largest seed funding rounds in the history of Silicon Valley, backed by investors including Andreessen Horowitz and Nvidia.

While Inkling shows promise, it still trails behind Chinese models like GLM 5.2 and Kimi K2.6 on major performance benchmarks. Notably, Inkling operates as a mixture-of-experts model, activating only parts of the network for specific inputs, allowing for efficient processing despite its large size. The model is pretrained on 45 trillion tokens and processes information across text, images, and audio, enabling a context window of 1 million tokens—equivalent to approximately 750,000 words.

Looking ahead, the AI landscape remains competitive, with companies vying for investments and advancements. Murati's Inkling could shift dynamics, especially for developers seeking accessible AI tools as it opens resources that have typically been proprietary. Watching how Inkling fares against established models and any future funding rounds for Thinking Machines Lab will be crucial for understanding the model's impact on the broader AI ecosystem.