Mira Murati unveils real-time AI models

• Mira Murati unveils AI models for real-time, natural conversations.
• Models use 200ms segments, respond under 0.4s, surpass Gemini.
• Public release later this year; improves user experience, support.

Imagine AI that understands and responds like a human, without the frustrating lag. Mira Murati's new models promise just that, offering fluid, real-time audio-video conversations that feel incredibly natural.

Revolutionary Interaction Models

Thinking Machines founder Mira Murati has unveiled a groundbreaking advancement in artificial intelligence: 'interaction models.' These innovative models are designed to process both spoken words and visual cues simultaneously, enabling AI to respond with unprecedented speed and fluidity. The core objective is to eliminate the awkward pauses and delays that often plague current AI interactions, making conversations feel as seamless and natural as chatting with another person. This is achieved through a sophisticated 'full-duplex' approach, which breaks down conversational exchanges into incredibly small, 200-millisecond segments. This meticulous segmentation allows for almost immediate feedback, drastically improving the user experience. One of these powerful models boasts a staggering 276 billion parameters, adept at managing rapid dialogue while concurrently handling more complex tasks such as in-depth reasoning or web searches, showcasing remarkable multitasking capabilities. This technology aims to bridge the gap between human intuition and AI responsiveness.

Unmatched Speed and Performance

The remarkable responsiveness of these new interaction models is largely attributed to a technique called 'encoder-free early fusion.' This innovative approach significantly accelerates the response time, enabling outputs in under 0.4 seconds. To put this into perspective, this speed surpasses that of established systems like Google's Gemini-3.1-flash-live when evaluated on the FD-bench benchmark for full-duplex conversations. This level of performance is crucial for applications where every millisecond counts. While currently accessible only to a select group of research partners, a wider public release is anticipated later this year. The potential applications are vast and could redefine user experiences across numerous sectors. Consider the implications for immediate healthcare alerts, where swift notifications can be critical, or for real-time customer support, where instant problem-solving can dramatically enhance satisfaction and efficiency. The leap in conversational speed marks a significant milestone in AI development.