Mira Murati's Thinking Machines Unveils 'Interaction Models' — a Full-Duplex Bet Against the Prompt-and-Response Paradigm
The ex-OpenAI CTO's new lab publicly launched its first research: multimodal models that listen, watch, think, and respond simultaneously — not in turns. The team argues that every AI system will eventually need something like this.
Thinking Machines, the AI lab founded by former OpenAI CTO Mira Murati, broke cover on Monday with its first major research release: a class of models the company calls "Interaction Models," designed to work with humans in continuous, full-duplex collaboration rather than the turn-based prompt-response loop that defines virtually every commercial AI product today. As @thinkymachines put it in its launch post: "People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way."
The announcement landed with notable firepower behind it. Murati herself framed the company's thesis in explicitly architectural terms, writing that "interactivity has to be in the model, and it has to scale with intelligence," as shared via @miramurati. This isn't a product wrapper around an existing foundation model — it's a claim that the interaction layer itself must be a first-class component of training, not an afterthought bolted on at inference time. The distinction matters: most current voice and multimodal systems simulate real-time interaction by chaining fast turn-taking, not by genuinely processing multiple modalities concurrently during a shared session.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.