Mira Murati's new venture, Thinking Machines Lab, has unveiled a new class of artificial intelligence models designed for real-time collaboration with humans, signalling what could be the beginning of the end for traditional “turn-based” AI interactions.
In a research preview announced on May 11, the startup introduced TML-Interaction-Small, a 276-billion-parameter Mixture-of-Experts (MoE) model with 12 billion active parameters that processes conversations as continuous live streams instead of stop-and-start exchanges.
The company said its interaction models are built from scratch to support simultaneous communication across audio, video and text, enabling users and AI systems to speak, listen, observe, search, draw and use tools at the same time during an ongoing interaction.
Unlike conventional chatbots that wait for users to finish before responding, the new system breaks communication into 200-millisecond “micro-turns”, allowing the AI to continuously process incoming information while generating replies. According to the company, the model can even follow timing-based instructions, such as delaying a response for a few seconds.
Thinking Machines Lab said existing AI systems are largely optimised for autonomous task completion, where users give instructions upfront and wait for results. The company argued that such “turn-based” systems create a collaboration bottleneck by limiting users' ability to interrupt, guide or clarify tasks in real time.
The startup also criticised the industry's dependence on external software “harnesses” for handling interruptions and multimodal interaction, saying interactivity should be embedded directly into the model architecture.
“We think interactivity should scale alongside intelligence,” the company said, adding that AI systems should collaborate with humans “the same way we do with other people.”
Essential Business Intelligence, Continuous LIVE TV, Sharp Market Insights, Practical Personal Finance Advice and Latest Stories — On NDTV Profit.
