Thinking Machines Lab Is Building AI That Talks Back

Mira Murati's Thinking Machines Lab has raised $2 billion and launched a new class of AI it calls "interaction models," systems capable of processing input and generating output simultaneously so they can be interrupted mid-response, the way a human can be cut off in conversation. The first model, TML-Interaction-Small, responds in 0.40 seconds. That is not a benchmark tweak. That is a fundamental rethinking of how humans and AI systems exchange information in real time.

Founded in February 2025, Thinking Machines Lab was built from the start with a specific thesis: that the most powerful AI systems are also the least usable, and that closing that gap is where the real competitive edge lives. The lab pulled significant talent from OpenAI when it launched, and Murati, as former CTO of the world's most closely watched AI company, brought with her both credibility and a precise understanding of where frontier model development has fallen short. The Tinker platform is now live for AI researchers, signaling the lab is serious about building an ecosystem, not just shipping demos.

The infrastructure backing here is not trivial. Nvidia has committed at least one gigawatt of computing power through its next-generation Vera Rubin chips under a multi-year strategic partnership, and is making an additional significant investment in the company. Google is also involved in infrastructure support. When two of the most powerful compute providers in the world simultaneously back the same early-stage AI lab, that is a signal worth reading carefully. It suggests both companies see Thinking Machines Lab as a serious contender in the next wave of model development, not just an ambitious side project.

For founders and investors, this story carries several layers. First, it confirms that talent mobility out of OpenAI is not a footnote but a structural force reshaping the frontier AI landscape. Second, the $2 billion raise at this stage signals that capital markets are still willing to bet heavily on pedigree combined with genuine technical differentiation. Third, the interaction model category itself, if it matures, could redefine expectations for AI interfaces across every vertical, from healthcare and education to enterprise software. Products built on top of conversational AI will need to rethink latency and interruption handling as table stakes.

In the next twelve months, the pressure on Thinking Machines Lab shifts from fundraising and research previews to demonstrable product traction. The Tinker platform and the interaction model research preview are credible opening moves, but the lab will need to show that its differentiated model architecture translates into developer adoption and real deployment at scale. Watch for enterprise partnerships, API pricing announcements, and whether any major product companies build on TML models in production. If they do, Murati will have done something genuinely rare: built a post-OpenAI company that competes with OpenAI on its own terms.