Thinking Machines describes an interaction model designed to keep receiving information while responding. Its research announcement divides incoming and outgoing audio, video, and text into short intervals, processing input and generating output in 200-millisecond chunks. This lets the conversation continue without waiting for an entire response to finish.
Longer reasoning and tool work go to a background model. The interaction model keeps taking input and incorporates results as they arrive, so a user can clarify the task while work continues. The chunk size describes the processing design; the proposed benefit is maintaining a responsive conversation alongside slower work.
