Mira Murati has newly open-sourced the Inkling large model and launched it on OpenRouter, focusing on enterprise-grade proxy tool calling.

PANews July 26 news: According to Decrypt, Thinking Machines Lab under former OpenAI CTO Mira Murati released its first open-source large language model Inkling on July 15. It uses a 975 billion-parameter MoE architecture, activates about 41 billion parameters during inference, supports text, image, and audio inputs, and a context window of around one million. In MCP Atlas agent tool-calling evaluations, Inkling achieved a 74.1% task completion rate, and in SWE-Bench Verified its completion rate reached 77.6%. Inkling is open-sourced on Hugging Face under the Apache 2.0 license, and has been deployed on OpenRouter. Pricing is $1 per million input tokens and $4.05 per million output tokens. It is suitable for agent workflows based on OpenRouter such as Hermes and OpenClaw, and mainly targets institutional users who cannot use China-made models and place importance on compliance and customizability.