Thinking Machines is developing a fully duplex AI capable of listening and speaking simultaneously

Thinking Machines Lab launches a new type of AI model

The startup founded last year by former OpenAI CTO Mira Murati announced its latest development—a conversational system that can interrupt a person during interaction.

What’s changed
* Traditional models

Modern AI models operate in “half‑duplex”: the user speaks → the model listens; when the model responds, the user is left only to listen. This feels more like messaging than a live conversation.

* Thinking Machines’ new approach

The company created a full‑duplex system that simultaneously processes incoming speech and generates a response. As a result, the interaction looks like a normal phone call where both parties can speak without pauses.

Technical details
* TML‑Interaction‑Small model

- Response time: 0.4 seconds

- This is almost twice as fast as comparable models from OpenAI and Google.

- The speed matches natural human speech.

How the technology will be available
* Currently it is a research version not intended for general public use.

* In the coming months the company plans to release a limited research version.

* It is expected that the broader audience will be able to use TML‑Interaction‑Small by the end of the year.

Thus, Thinking Machines Lab offers a new format of AI communication that feels closer to a real conversation and significantly speeds up information exchange.