Thinking Machines shows AI that responds in real time using voice and video

In short: Thinking Machines, founded by former OpenAI CTO Mira Murati, says it is building “interaction models” that can take in voice, video, and text at the same time and respond as things happen.

What happened

Thinking Machines announced it is working on what it calls “interaction models.” The company says these systems are meant to work more like a back and forth conversation with a person, instead of the usual pattern where you talk, wait, then get an answer.

The company argues that many chatbots act like they are stuck in a single line of messages. For example, they often do not react until you stop speaking or finish typing, and they typically stop “paying attention” while they are generating a reply. Thinking Machines says its approach aims to remove that bottleneck, like switching from email to an in-person conversation.

Thinking Machines shared demo examples to show what it means. In videos, the AI listens to a story and flags when animals are mentioned, translates speech while someone is still talking, and uses video to tell someone when they are slouching (sitting with poor posture).

The company says people cannot try these interaction models yet. It plans a “limited research preview” in the coming months, with a wider release later this year.

Thinking Machines was founded in February 2025. The Verge notes the lab has also dealt with high profile staff departures to other companies.

Why it matters

If this works as described, AI tools could feel less like a form you fill out and more like a helper that stays with you while you work, talk, or move around. That could make features like live translation and real-time coaching more useful, but it also raises practical questions about when a tool is listening or watching and how that data is handled.

Source: The Verge AI