Mira Murati’s Thinking Machines Lab Introduces Interaction Models: A Native Multimodal Architecture for Real-Time Human-AI Collaboration

Thinking Machines Lab unveiled a research preview of interaction models for real-time multimodal AI.

The system handles audio, video, and text in continuous micro-turns instead of fixed turn-based exchanges.

A real-time interaction model works alongside a separate background reasoning model for deeper asynchronous tasks.

The design aims to keep conversation flowing while enabling richer collaboration, faster responses, and ongoing tool use.