Thinking Machines, the artificial intelligence startup founded by former OpenAI Chief Technology Officer Mira Murati, announced Monday the development of "interaction models" designed to enable real-time conversation with AI systems. The company demonstrated technology that processes audio, video, and other inputs simultaneously rather than waiting for users to finish speaking before responding.

The startup's approach represents a departure from current AI systems that operate in a turn-based format, where users provide input and wait for the AI to process and respond. Thinking Machines' new system uses what it calls a "full-duplex" architecture that processes 200-millisecond chunks of input and output at the same time, allowing the AI to listen, speak, and analyze visual information concurrently.

The research preview features TML-Interaction-Small, a 276-billion parameter model with 12 billion active parameters. According to performance tests, the system achieved a response latency of 0.40 seconds compared to 1.18 seconds for OpenAI's GPT real-time system and 0.57 seconds for Google's Gemini live model. The company said the technology scored 77.8 on interaction quality benchmarks, significantly higher than competing systems.

Thinking Machines was founded in 2024 by Murati along with former OpenAI researcher John Schulman and other former OpenAI staff. The company raised approximately $2 billion in funding at a $12 billion valuation in a round led by Andreessen Horowitz, with participation from Nvidia, Accel, and other investors. The startup has experienced some employee turnover, with several founding members joining Meta, though it has also recruited high-profile talent including PyTorch creator Soumith Chintala as CTO.

The interaction models are not yet available to the public or enterprise customers. Thinking Machines said it plans to open a limited research preview in the coming months to collect feedback, with a wider release planned for later this year. The company has not disclosed whether the technology will be released as open-source software, following its previous commitment to significant open-source components in its releases.