According to CoinMeta, Thinking Machines Lab, founded by CTO Mira Murati before OpenAI, has released its first general-purpose model, Inkling. The model has a total of 975 billion parameters, with 41 billion parameters activated each time, and it supports text, images, and audio. The maximum context length is 1 million token. Inkling was trained from scratch, but its MOE architecture mainly refers to DeepSeek-V3. The complete weights have been made available on Hugging Face. Thinking Machines Lab has made Inkling and Inkling-small available for free on OpenRouter. The testing is expected to last for several weeks and is only available for agentic and harnesses users. The free endpoints will record users' prompt, model outputs, and usage data, which will be used to improve the model, products, and services after being unbound from long-term identity identifiers such as account information.