Inkling is Thinking Machines Lab’s first proprietary AI model and its first public proof point since the company emerged from stealth. It’s a mixture-of-experts system with 975 billion total parameters (about 41 billion active per task), trained on 45 trillion tokens of text, image, and audio, with a context window up to 1 million tokens. Unlike the closed, general-purpose chatbots sold by OpenAI, Anthropic, and Google, Inkling ships open-weight under Apache 2.0, and supports a “controllable thinking effort” setting that lets users trade reasoning depth for speed and token cost.
The company, founded by former OpenAI CTO Mira Murati, is positioning Inkling less as a finished chatbot than as a starting point for enterprises to fine-tune themselves through Tinker, Thinking Machines’ model-customization platform — the bet being that organizations willing to own and adapt their own models can extract more value than by renting access to a one-size-fits-all frontier model. Inkling was trained entirely on Nvidia GB300 NVL72 systems under a strategic infrastructure partnership announced in March 2026.