Market intelligence

Thinking Machines Unveils 'Interaction Models' for Real-Time AI in May 2026

Thinking Machines, the AI lab from former OpenAI CTO Mira Murati, revealed its work on 'interaction models.' This new architecture enables AI to process audio, video, and text simultaneously in real time, aiming to create more natural human-AI collaboration.

2 min read

Key takeaways

  1. 01Thinking Machines, led by former OpenAI CTO Mira Murati, has unveiled 'interaction models,' a new AI architecture for real-time, multimodal human-AI collaboration.
  2. 02The technology aims to solve the 'bandwidth bottleneck' of current models, which process information sequentially rather than continuously.
  3. 03The company demonstrated use cases like real-time translation and posture correction with a model named TML-Interaction-Small.
  4. 04A limited research preview is planned for the coming months, with a wider release anticipated later this year.

A New ParadigmLede

Mira Murati's AI company, Thinking Machines, announced on Monday its development of a new class of AI called 'interaction models.'[1] The technology is designed to enable continuous, real-time collaboration between humans and AI systems across multiple modalities, including audio, video, and text.

Solving the BottleneckEvent Summary

The company argues that current AI models suffer from what it calls a narrow channel for human-AI collaboration. These systems typically wait for a user to finish an input before processing, and their own perception is frozen while generating a response. Thinking Machines' objective is to resolve this bandwidth bottleneck by creating AI that is interactive in real time.

The new models are reportedly trained from the ground up on continuous data streams. This allows the AI to perceive and react concurrently, much like human interaction. The company demonstrated several use cases, including an AI that translates speech in real time and another that detects a user's posture.

Murati's VisionPublisher Context

Thinking Machines was founded by Mira Murati in February 2025 after her departure from her role as Chief Technology Officer at OpenAI. The move signaled a new direction for the prominent AI executive.

The AI research lab has reportedly faced challenges since its inception, including the departure of key staff members to competitors like Meta, with some even returning to OpenAI. This announcement marks the first major public reveal of the company's core research, positioning it as a direct challenger to the prevailing agent-focused direction of other major AI labs.

Path to ReleaseOutlook

The technology, demonstrated with a model named TML-Interaction-Small, is not yet available for public use.[2] Thinking Machines has stated its intention to open a limited research preview in the coming months.

A wider release later this year is the stated goal, though specific dates have not been provided. This phased rollout suggests a strategy focused on gathering feedback from a controlled group of researchers before a broader deployment. The success of this preview will likely determine the timeline and scope of the eventual public offering.

A Fundamental ShiftWrapup

The introduction of interaction models represents a potential shift in the trajectory of AI development. Instead of focusing solely on creating autonomous agents, Thinking Machines is emphasizing the quality of the human-computer interface. The core principle is that responsiveness and intelligence should be trained together, not treated as separate components.

[3]By building interactivity into the model itself, Murati's lab challenges the fundamental architecture of today's AI systems. If successful, this approach could lead to more intuitive and powerful collaborative tools, fundamentally changing how people work with artificial intelligence.

Citations

  1. [1]

    Mira Murati's AI company, Thinking Machines, announced on Monday its development of a new class of AI called 'interaction models.'

    "Thinking Machines, the AI company founded by former OpenAI CTO Mira Murati, announced Monday that it’s working on something called “interaction models.”"
  2. [2]

    The technology, demonstrated with a model named TML-Interaction-Small, is not yet available for public use.

    "The model, named TML-Interaction-Small, processes audio, video, and text simultaneously and in real time"
  3. [3]

    The core principle is that responsiveness and intelligence should be trained together, not treated as separate components.

    "The core principle is that responsiveness and intelligence should be trained together from scratch, on continuous audio and video streams"

Sources

4 references

Maxime Doussin, CTO at MWM

Maxime Doussin

CTO

Maxime Doussin is the CTO of MWM, where he leads engineering, data infrastructure, and the mobile-app market-intelligence platform. He writes MWM's weekly app trend analysis, drawing on proprietary ranking data covering millions of iOS and Android apps across 150+ countries.

This article is an independent editorial analysis. App names, trademarks, and brands mentioned are the property of their respective owners. Market data and rankings referenced are based on MWM's proprietary estimates.

Believe this article infringes your intellectual property? File a dispute