Mira Murati, the former chief technology officer of OpenAI, has announced that her new AI startup Thinking Machines is developing what it calls interaction models, a new category of AI system designed for simultaneous, real-time processing of audio, video, and text rather than the sequential chat-based exchanges that define most current AI interfaces. The announcement marks one of the most significant public updates from Thinking Machines since its founding and positions the company as a direct challenger to the conversational AI paradigm that has defined the industry since the launch of ChatGPT.
What Happened
Thinking Machines has been working on AI systems that process audio, video, and text simultaneously rather than waiting for a user to complete their input before generating a response. Unlike traditional large language models that receive a prompt and generate a response in sequence, the interaction models demonstrated by Thinking Machines interpret all three modalities in real time, meaning the system is continuously aware of what the user is saying, showing, and typing at the same moment.
Demonstrations of the technology included live translation between speakers in different languages with near-zero perceptible lag, contextual awareness that allowed the system to respond to gestures and environmental cues visible on camera, posture monitoring that triggered context-sensitive responses based on the user physical state, and a dynamic conversation style that adapted in real time rather than waiting for turn-based exchanges. These capabilities suggest a system architecture significantly different from transformer-based chat models, though Thinking Machines has not disclosed technical details of the underlying approach.
Murati described the goal as enabling more natural collaboration between humans and AI systems, arguing that the turn-taking format of current AI interfaces imposes an unnatural constraint on how people can work with AI. Interaction models are designed to remove that constraint and allow the AI to be a continuous, responsive presence rather than a tool you query.
Why It Matters
The interaction model approach, if it scales, would represent a meaningful departure from how frontier AI systems currently work. The dominant paradigm in AI interfaces is still fundamentally chat-based, even when wrapped in voice or video interfaces, because the underlying model processes inputs sequentially. Building a system that is genuinely multimodal and real-time at the model level, rather than as a surface-level interface on top of a text model, is a significantly harder technical challenge.
Thinking Machines has not announced a release timeline, product availability, or pricing. The announcement appears designed to establish the company research direction and competitive positioning ahead of a product launch. Given Murati track record at OpenAI, where she oversaw the development and release of GPT-4, DALL-E 3, and Sora, the announcement carries credibility that a typical startup claim would not. The AI industry will be watching closely to see whether the interaction model demonstrations represent a genuinely novel capability or a well-staged preview of technology that still has significant engineering work ahead of it.
What Comes Next
Thinking Machines has not disclosed its funding status, team size, or compute infrastructure. The company is one of several high-profile AI startups founded by senior alumni of major AI labs, a cohort that includes companies from former Google, Meta, and OpenAI researchers. Interaction models represent a compelling differentiation thesis in a market that is increasingly crowded at the chat-model layer, and the coming months will reveal whether Thinking Machines can translate that thesis into a working product.
Stay updated on the latest AI news at Evolve Digital.