Mira Murati's Open-Source Inkling Lands at 975B Parameters—AI's Biggest Brain Dump Yet 🧠

Mira Murati's Thinking Machines Lab released its first AI model this week, Inkling, trained from scratch and published in full under an Apache 2.0 license with weights available on Hugging Face. The release marks the company's first public product since Murati founded Thinking Machines Lab in February 2025, roughly 10 months after she departed OpenAI, where she had served as CTO and briefly as interim CEO during Sam Altman's November 2023 ouster and five-day reinstatement.

Inkling is a multimodal mixture-of-experts model totaling 975 billion parameters with 41 billion active per inference, accepting text, images, and audio through a 1 million-token context window. It was pretrained on 45 trillion tokens spanning text, images, audio, and video. Murati announced the model on X: "Our first model, Inkling. Trained from scratch, weights are open, fine-tunable on Tinker today." Tinker is Thinking Machines' cloud platform aimed at fine-tuning workflows.

The company had raised $2 billion at a $12 billion valuation in July 2025 in a round led by Andreessen Horowitz with Nvidia, Accel, ServiceNow, Cisco, AMD, and Jane Street participating—one of the largest seed rounds in Silicon Valley history at the time. Reports in November 2025 indicated the firm was pursuing a new round at a $50 billion valuation; those talks had collapsed by January 2026.

On agentic benchmarks, Inkling posted 74.1% on MCP Atlas, nearly 30 percentage points above Nvidia's Nemotron 3 Ultra, and 77.6% on SWE-Bench Verified, compared with Nemotron's 70.7%. Thinking Machines is positioning the model as "well-rounded" across reasoning and tool-use tasks, with full weights downloadable at no cost and no usage restrictions under the Apache 2.0 terms.