Thinking Machines Lab, founded by Mira Murati, announced on Monday, May 11, what it's calling "interaction models," a new type of AI built to handle conversations, video, and collaboration in real time rather than in turns.
The startup, which is reportedly in talks for funding at around a $50 billion valuation, has shown demos of AI that listens while it talks, processes what it sees in video while responding, and jumps into conversations the way a person would, rather than waiting for you to finish every sentence.
What Murati sees as a core problem is what the company's approach is addressing: "Interactivity should scale alongside intelligence," the company noted in its official post, arguing that the way we work with AI shouldn't be an "afterthought."
Its flagship model, TML-Interaction-Small, responds in 0.40 seconds and handles audio, video, and text at the same time. By comparison, Google's Gemini-3.1-flash-live takes 0.57 seconds, while OpenAI's GPT-realtime-2.0 takes 1.18 seconds.
What Thinking Machines is actually building
The focus isn't just faster responses. Thinking Machines is trying to solve what it calls the "bandwidth bottleneck" between humans and AI. Right now, you type or speak, the AI waits, processes everything, and then responds. That back-and-forth limits how much of your knowledge and intent can reach the system.
Thinking Machines Lab built its models from scratch to work continuously instead. The system breaks interaction into 200-millisecond chunks, processing what you're saying and showing while also thinking and responding. One part handles the conversation flow while another works on complex tasks in the background, similar to how humans can talk while thinking about what to say next.
Demos shared alongside the announcement on Monday showed the AI counting exercise repetitions from video, translating speech in real time, and noticing when someone slouches - all while maintaining a conversation.
The goal is to create a form of collaboration that feels more like working with another person than sending requests to a machine.
The OpenAI exodus backstory
Murati built her credibility on this vision at OpenAI, where she served as CTO and led the work behind ChatGPT. She briefly became interim CEO during the board drama in November 2023 when Sam Altman was fired, then reinstated days later. Ten months after Altman's return, she left OpenAI in September 2024 and founded Thinking Machines Lab in February 2025.
Monday's announcement also reflects the talent battles that have shaped the company's first year. Meta reportedly tried to buy Thinking Machines in 2025 and, when Murati declined, hired away seven founding members. Murati responded by bringing in Soumith Chintala, the creator of PyTorch, as CTO.
When you can actually use it
The technology isn't available yet. This is a research preview, with Thinking Machines saying a limited version is coming to research partners in the next few months and a wider public release later this year.
From the Demos, we see AI that can watch, listen, and respond at the same time, bringing the company's vision of continuous human-AI collaboration closer to the conversational experience people have been expecting since the movie "Her." For a startup that's just 15 months old, Monday's announcement marks the first major public reveal of what Murati has been building since leaving OpenAI, and why investors have reportedly been willing to value it at $50 billion.
