Mira Murati's AI Startup Is Building a Model That Can Interrupt You Mid-Sentence

[madmax]

Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has announced a new category of AI it calls interaction models — systems designed to process your input and generate a response simultaneously rather than waiting for you to finish speaking before beginning to reply. The technical term for this architecture is full duplex communication, and it represents a fundamental departure from how every current AI model works. Today's AI systems operate in a strict turn-taking sequence: you speak, it listens, it responds, you listen. That structure more closely resembles exchanging text messages than having an actual conversation, and it is one of the primary reasons that even the most capable AI voice assistants feel slightly unnatural compared to talking with a person. Thinking Machines' model, called TML-Interaction-Small, claims a response latency of 0.40 seconds — roughly the speed of natural human conversation and significantly faster than comparable models from OpenAI and Google according to the company's benchmarks. The underlying design philosophy is that interactivity should be native to a model's architecture rather than layered on top of a system that was not built for it, which is how most current voice AI implementations work.

The announcement is a research preview rather than a product launch, meaning the public cannot access TML-Interaction-Small yet. A limited research preview is planned for the coming months with a wider release later this year. The honest assessment at this stage is that the benchmarks are impressive and the concept is genuinely interesting — a model that can participate in conversation the way a human does, including the ability to interrupt, backtrack, and respond to mid-sentence changes in direction, would represent a meaningful step toward AI that feels like a collaborator rather than a sophisticated answering machine. Whether the real-world experience matches the technical claims is something that will only become clear once researchers and eventually the public can actually use it. Thinking Machines Lab has positioned itself as a serious research organization since its founding, and Murati's credibility from her time at OpenAI gives the announcement more weight than it would carry from a less established team — but interaction models remain an unproven category, and the gap between an impressive benchmark and a genuinely useful conversational AI product has surprised the industry before.

[tradelikepro]

0.40 second latency and i still say "um" for 3 seconds before my actual question, the bottleneck is me apparently

[bonk]

Mira left OpenAI and built what they didn't

[bonk]

0.40 seconds, finally faster than my thoughts