Mira Murati's Thinking Machines pursues continuous multimodal AI collaboration
Mira Murati's startup announced Monday it is building interaction models that process audio and video continuously, aiming to replicate natural human collaboration patterns rather than discrete prompt-response cycles.
Thinking Machines, the AI company founded by former OpenAI CTO Mira Murati, announced Monday that it's working on "interaction models" — systems designed to let people "collaborate with AI the way we naturally collaborate with each other" by continuously ingesting audio, video, and other inputs rather than processing discrete prompts.
The announcement marks the first public disclosure of Thinking Machines' technical direction since Murati departed OpenAI in September 2024. During her six-year tenure as CTO (2018–2024), she oversaw development of GPT-4, DALL-E 3, and the initial ChatGPT launch. She also briefly held the CEO role during the November 2023 board crisis that temporarily ousted Sam Altman.
The "interaction model" concept suggests a system that maintains context across multiple sensory channels simultaneously, treating AI collaboration as a persistent multimodal stream rather than a turn-based exchange. That approach would align with recent industry momentum toward real-time multimodal systems, though most current implementations still operate on request-response cycles even when handling audio or video.
Thinking Machines did not share a release timeline, model architecture details, benchmark numbers, parameter counts, or demo videos. The company also did not disclose whether the interaction models will be open-weight or API-only, leaving open questions about how they differ architecturally from existing multimodal transformers and whether the continuous-input design requires new training regimes or can be achieved through inference-time changes to existing model families.
Murati's startup follows a pattern of senior OpenAI departures over the past eighteen months. Chief Scientist Ilya Sutskever left in May 2024 to found Safe Superintelligence Inc., and several other research leads have launched their own ventures. Thinking Machines has not disclosed funding details, team size, or whether it plans to compete directly with OpenAI's product suite or focus on a different market segment.