Thinking Machines Lab previews interaction models

Thinking Machines Lab, the startup founded by former OpenAI chief technology officer Mira Murati, previewed a set of voice-based systems it called “interaction models.” Where conventional voice assistants wait for a user to finish speaking before responding, Semafor reported that the new models were designed to interrupt and inject context mid-conversation, aiming for the back-and-forth rhythm of a human exchange rather than a strict turn-taking format.

The announcement was a research preview rather than a public release: the company gave few technical details about the underlying architecture, and Semafor reported that a wider rollout was planned later in 2026. Thinking Machines had raised roughly $2 billion in venture funding by this point and had shipped little in the way of public products since its 2024 founding, making the preview one of its first concrete signals of what it intended to build beyond training infrastructure and tooling. The framing — conversational latency and interruption as a differentiator — positioned the lab against both OpenAI’s and Google’s existing voice assistants, though no benchmark or user comparison accompanied the preview.