Mira Murati's Thinking Machines Lab Releases Inkling, a 975B-Parameter Open-Weight AI Model
The Western-developed model targets enterprise customization and agentic workflows with Apache 2.0 licensing and strong MCP Atlas performance.
Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, released Inkling on July 15, 2026—a 975-billion-parameter open-weight model positioned as a customizable foundation for enterprise deployments rather than a raw performance leader.
Architecture and Performance
Inkling uses a Mixture-of-Experts architecture with 975 billion total parameters and 41 billion active parameters per task. The model was pretrained from scratch on 45 trillion tokens spanning text, images, audio, and video, and supports a 1-million-token context window.
On the MCP Atlas benchmark for agentic tool use, Inkling scored 74.1%, outperforming Nvidia's Nemotron 3 Ultra at approximately 44.7%. This positions the model as a strong option for developers building autonomous pipelines and tool-using agents.
Thinking Machines Lab stated in its announcement: "Inkling is not the strongest overall model available today, open or closed." The company is betting against one-size-fits-all AI, emphasizing customization and compliance-friendly deployment over leaderboard dominance.
Licensing and Access
Inkling is available on Hugging Face under the Apache 2.0 license, permitting commercial use and modification without royalty obligations. The model is also accessible via OpenRouter, with pricing set at $1 per million input tokens and $4.05 per million output tokens.
The permissive license and Western origin address a strategic gap: compliance-driven organizations unable or unwilling to use Chinese open-weight models like Qwen or GLM now have a high-capacity alternative developed outside China.
Market Context
Thinking Machines Lab was founded in September 2024 following Murati's departure from OpenAI. The company is valued at $12 billion, according to Fortune. Inkling marks the lab's first major model release, arriving as Western laboratories have struggled to match the pace of Chinese open-weights development.
A Decrypt reviewer characterized Inkling as "the best open-source model a Western lab has shipped—and that is both its main selling point and its ceiling." The assessment reflects the model's positioning: competitive in agentic tasks and enterprise customization, though not leading in raw coding benchmarks or cost-efficiency compared to smaller specialized models.
Enterprise Positioning
The model's design priorities—agentic tool use, multimodal pretraining, and a permissive license—target organizations building production AI pipelines that require auditability, customization, and regulatory compliance. The 1-million-token context window enables processing of lengthy documents, codebases, and multimodal inputs without truncation.
Inkling's release signals Thinking Machines Lab's commitment to an open-weight strategy distinct from the closed-model approaches dominant among major U.S. AI labs. The bet: enterprise buyers will value compliance and customization over raw benchmark supremacy.