Thinking Machines wants to build an AI that actually listens while it talks
By Jakub Antkiewicz
•2026-05-12T10:32:02Z
Mira Murati's Startup Aims for More Natural AI Conversation
Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has announced a new architectural approach called interaction models. The initiative aims to shift AI assistants away from the current turn-based, text-chain-like cadence toward a more fluid, real-time conversational style. The core objective is to build a model that can process a user's speech and generate a response concurrently, enabling it to interrupt and interact in a manner more akin to a natural human phone call.
Technical Details and Performance Claims
The technical foundation for this capability is known as “full duplex” communication. Thinking Machines Lab claims its first demonstration model, TML-Interaction-Small, achieves a response time of 0.40 seconds, a latency that closely mirrors natural human conversational speeds. This is presented as a significant improvement over comparable models from competitors like OpenAI and Google. The company has not yet released the model publicly, labeling it a research preview.
- Model: TML-Interaction-Small
- Technology: Full duplex, simultaneous processing and generation
- Claimed Response Time: 0.40 seconds
- Status: Research preview, with a limited release planned in the coming months.
The release strategy involves a limited research preview in the near future, followed by a wider release later in 2026. While the company's benchmarks are notable, the actual user experience remains hypothetical until the model is accessible. The success of this approach will depend entirely on whether the real-world performance aligns with the technical specifications and if users find the interactive style more effective than current systems.
By prioritizing low-latency, full-duplex interaction, Thinking Machines is signaling a strategic bet that the qualitative feel of a conversation, not just the accuracy of the response, will become a key competitive differentiator in the AI market.