Thinking Machines unveils real-time AI interaction models to tackle human-AI bandwidth limits
The company plans a limited research preview in the coming months, with a wider release targeted for later this year, as it seeks to move beyond single-threaded processing
Thinking Machines has announced the development of "interaction models" designed to facilitate real-time collaboration between humans and artificial intelligence. Founded by former OpenAI Chief Technology Officer Mira Murati in February 2025, the company aims to overcome current bandwidth bottlenecks by allowing systems to continuously perceive and act upon audio, video, and text inputs. This approach contrasts with traditional models that operate in a single thread, freezing their perception of the user until input generation is complete.
The new technology enables a more natural flow of interaction by allowing the model to process user actions and intent without interruption. By resolving the limitations of current single-threaded architectures, Thinking Machines intends to ensure that the full scope of human knowledge, intent, and judgement can be utilised effectively. The company argues that this shift allows AI interfaces to meet users where they are, rather than forcing humans to adapt their behaviour to fit existing interface constraints.
In demonstrating early capabilities, the company highlighted features such as real-time speech translation, the ability to detect mentions of animals within a narrative, and posture monitoring that alerts users when they are slouching. These examples illustrate the potential for the system to engage with users dynamically across various modalities. However, the specific technical architecture and precise latency requirements of the interaction models have not been detailed beyond this high-level description.
Thinking Machines plans to open a limited research preview of the technology in the coming months, with a wider public release targeted for later this year. While the company has outlined the strategic direction and intended benefits of the new models, exact release dates remain unspecified. The efficacy of the demonstrated features, particularly in real-world scenarios outside of internal demonstrations, also remains unverified at this stage.
The launch of Thinking Machines follows significant movement within the broader AI sector, with the company founded by Murati after her departure from OpenAI. Since its inception, the lab has experienced notable staff turnover, with key members reportedly defecting to Meta and returning to OpenAI. Despite these internal shifts, the company continues to push forward with its vision for more responsive and integrated AI systems.
