Mira Murati Launches Inkling: A New Open-Source AI Model

Mira Murati’s Thinking Machines Lab has unveiled Inkling, its first open-source AI model since she departed OpenAI in September 2024. Released on July 15, 2026, Inkling features 975 billion parameters and employs a mixture-of-experts architecture, where only 41 billion parameters are active during any single inference. This setup allows it to read and process text, images, and audio, and it supports a context window of up to one million tokens.

Inkling is available on the OpenRouter platform for a cost of $1 per million input tokens and $4.05 per million output tokens. While it boasts impressive MCP Atlas scores—74.1% completion rate for real-world tasks and 77.6% on the SWE-Bench Verified coding performance—it faces stiff competition. Competing models, including those from Chinese labs, typically offer better performance for similar or lower costs, which raises questions about Inkling’s pricing strategy.

The launch comes at a time when Western labs are struggling to keep up in the open-source AI race. Previously, models from companies like Alibaba and Z.ai have dominated the leaderboard, making Inkling's arrival noteworthy. However, it remains to be seen how well it can perform in practical applications compared to its competitors.

Mira Murati notes, "Inkling represents a significant step for us, positioned to redefine the landscape of open-source AI despite the existing competition." This statement underscores the ambition behind Inkling, aiming to make inroads in an increasingly competitive sector.

As the AI landscape evolves, industry watchers will keep an eye on its adoption and performance metrics, particularly how it stacks up against rivals in terms of cost-efficiency and output quality.