Thinking Machines Launches Interaction Model with 200ms Response, Outperforms GPT-Realtime-2.0

According to Beating, Thinking Machines, the lab founded by former OpenAI CTO Mira Murati, released a research preview of its Interaction model, featuring native real-time audio and video processing with 200-millisecond micro-turn responses. The model enables simultaneous listening, viewing, and speaking while supporting real-time user interruptions.

The TML-Interaction-Small model uses a 276-billion-parameter MoE architecture with 12 billion parameters activated per inference. Official data shows a speech turn-taking latency of 0.40 seconds and a FD-bench V1.5 score of 77.8, both exceeding GPT-Realtime-2.0 and Gemini 3.1 Flash Live. Limited preview access is planned for the coming months.