What are Thinking Machines Interaction Models?

On May 11, 2026, former OpenAI CTO Mira Murati launched Thinking Machines Labs with a groundbreaking Interaction Model technology. Unlike traditional AI that operates in a "turn-taking" manner, these models use a full-duplex architecture that processes input and output simultaneously — the AI can listen, see, and respond in real-time, just like human conversation [citation:1][citation:9].

Key Features

  • Full-Duplex Architecture: AI processes input and output simultaneously (200ms chunks), eliminating the "wait and respond" bottleneck [citation:9].
  • Real-Time Multimodal Processing: Simultaneously handles audio, video, and text inputs without pre-processing [citation:1].
  • Dual-Model System: 276B parameter MoE frontend (12B active) handles real-time interaction; background model handles complex reasoning tasks asynchronously [citation:9].
  • 0.4s Latency: Nearly three times faster than GPT-realtime-2.0 (1.18s) [citation:9].
  • Natural Backchanneling: AI can nod or say "hmm" while you talk — a first for AI assistants.
  • Proactive Interruption: AI can interject when it spots errors (e.g., bugs in your code).

Benchmark Performance

  • FD-bench (Interaction Quality): 77.8 vs GPT-realtime-2.0 at 46.8
  • Latency: 0.40 seconds vs 1.18 seconds

Pricing & Availability

Status: Limited research preview expected in "coming months" with wider release later in 2026 [citation:1].

The company has raised approximately $2 billion at a $12 billion valuation from a16z and other investors [citation:9].

Pros

  • Revolutionary full-duplex interaction paradigm
  • Dramatically lower latency than any competitor
  • Led by the original GPT-4o architect
  • Truly natural conversation flow with backchanneling
  • Visual real-time interaction capability

Cons

  • Not yet publicly available
  • Significant talent attrition reported (staff moving to Meta/OpenAI)
  • Requires powerful hardware for deployment
  • Company is small (~130 people) relative to ambition

Who Should Use It?

Perfect for: Developers building next-generation voice agents, real-time assistants, and applications requiring natural human-AI collaboration.

Verdict

Mira Murati has delivered exactly what she promised — a true departure from the ChatGPT-era dialogue box. Interaction Models represent the second generation of AI user interfaces, moving from turn-taking to simultaneous collaboration [citation:9].

Rating: 4.7/5 - A paradigm shift in human-AI interaction.