Thinking Machines Unveils Real-Time Full-Duplex AI Interaction Models
Thinking Machines, founded by former OpenAI CTO Mira Murati, has unveiled new "interaction models" enabling real-time, full-duplex AI communication. Unlike traditional turn-based systems, these models process audio, video, and text simultaneously, allowing for natural, low-latency conversations akin to human phone calls. The initial model, TML-Interaction-Small, boasts a 0.40-second response time. Currently in research preview, the technology aims for a limited release in coming months and a wider launch in 2026, positioning the startup as a significant competitor in the evolving artificial intelligence landscape.
Editorial summary awaiting refresh
Cross-source coverage
Wire timeline
Mira Murati’s Thinking Machines Lab Prioritizes Human-AI Collaboration Over Automation
Mira Murati, former CTO of OpenAI and founder of Thinking Machines Lab, outlines her vision for artificial intelligence that prioritizes human collaboration rather than job replacement. In an interview with WIRED, Murati argues that keeping humans in the loop is essential for ensuring positive futures as superintelligence develops. Her startup recently previewed new 'interaction models' capable of understanding continuous, non-linear human communication through audio and video inputs, allowing for more natural and adaptive interactions compared to standard chatbot interfaces. This approach contrasts with major tech companies like OpenAI and Google, which focus on autonomous task completion. Thinking Machines Lab, which has raised billions in funding, previously launched Tinker, an API for fine-tuning open-source models. The company aims to amplify human intent and values through AI systems that perceive context and adapt in real-time. This strategy aligns with growing economic calls for AI development focused on human empowerment. While the new interaction models have not yet been publicly released, demonstrations highlight their potential to handle interruptions, tone changes, and subject shifts seamlessly, marking a significant shift toward personalized, human-centric AI infrastructure.
WIREDThinking Machines Lab Unveils Full-Duplex AI for Natural Voice Conversations
Thinking Machines Lab, a startup founded by former OpenAI CTO Mira Murati, has announced the development of full-duplex AI technology designed to eliminate awkward pauses in voice interactions. Unlike traditional systems that wait for users to finish speaking before processing, this new model, TML-Interaction-Small, can listen and generate responses simultaneously. The company claims a response time of 0.40 seconds, closely mimicking the rhythm of natural human conversation and surpassing current offerings from competitors like OpenAI and Google. This advancement aims to make AI assistants feel less artificial and more like participants in a phone call rather than a walkie-talkie exchange. However, the technology is currently in a research preview phase, with limited access expected in the coming months and a broader release planned for later in 2026. While the speed promises smoother interactions, experts note that timing management is critical to prevent interruptions or misunderstandings. The ultimate success of this innovation depends on real-world user testing to determine if faster responses genuinely enhance usability in everyday scenarios. Key details regarding pricing, platform support, and widespread availability remain undisclosed as the lab prepares for its initial rollout.
Digital TrendsMira Murati's Thinking Machines Unveils Interactive AI System
Mira Murati, the former Chief Technology Officer of OpenAI, has launched a new artificial intelligence system through her startup, Thinking Machines. Established in February 2025 following her departure from OpenAI after a leadership dispute with CEO Sam Altman, the company has rapidly attracted significant investor interest, seeking a valuation exceeding $50 billion. The newly revealed AI system is designed for real-time, natural interaction, functioning as a virtual companion capable of analyzing audio, video, and text inputs without traditional prompts. Demonstrations highlight its ability to engage in asymmetric conversations and respond to environmental contexts, drawing comparisons to the interactive AI depicted in the film Her. Murati emphasized that this model is trained from scratch for native real-time interaction, distinguishing it from existing turn-based systems like ChatGPT or Claude. This development marks a significant step in Murati's post-OpenAI career, positioning Thinking Machines as a potential competitor in the advanced AI landscape by focusing on seamless human-AI collaboration.
India Today | Latest StoriesThinking Machines Lab Unveils Full-Duplex AI Model for Real-Time Conversation
Thinking Machines Lab, an AI startup founded by former OpenAI CTO Mira Murati, has announced a new class of 'interaction models' designed to enable full-duplex communication. Unlike current AI systems that operate in a turn-based manner where users must wait for the model to finish processing before responding, this new technology allows the AI to process input and generate responses simultaneously. This approach aims to mimic the natural flow of human phone conversations rather than text messaging. The company claims its initial model, TML-Interaction-Small, achieves a response latency of 0.40 seconds, significantly faster than comparable offerings from competitors like OpenAI and Google. Currently available only as a research preview, the model is not yet public. Thinking Machines plans to release a limited research preview in the coming months, with a wider release scheduled for later in 2026. While benchmarks appear impressive, real-world performance remains to be seen as the technology moves from development to practical application.
TechCrunchThinking Machines Unveils Real-Time AI Interaction Models
Thinking Machines, an AI startup founded by former OpenAI executives Mira Murati and John Schulman, has announced a research preview of new 'interaction models.' These native multimodal systems aim to replace traditional turn-based chat with fluid, near-real-time voice and video conversations. The technology utilizes a 'full-duplex' architecture that processes input and output simultaneously in 200ms chunks, allowing the AI to listen, speak, and observe visual cues concurrently. To balance speed with complex reasoning, the system employs a dual-model approach: an Interaction Model for immediate dialog management and a Background Model for asynchronous tasks like web browsing or deep analysis. The preview features TML-Interaction-Small, a 276-billion parameter Mixture-of-Experts model. While not yet publicly available, the company plans a limited research release in the coming months, citing impressive performance on the FD-bench benchmark for interaction quality. This development marks a significant shift toward treating interactivity as a core component of model architecture rather than an external software layer.
VentureBeatMira Murati’s Thinking Machines Unveils Real-Time AI Interaction Models
Thinking Machines, the artificial intelligence startup founded by former OpenAI Chief Technology Officer Mira Murati, has announced the development of new 'interaction models.' These models are designed to enable real-time collaboration between humans and AI by continuously processing audio, video, and text inputs simultaneously. Unlike current AI systems that operate in a single thread and wait for user input to complete before responding, these interaction models aim to eliminate bandwidth bottlenecks by perceiving and acting in real time. The company argues this approach allows for more natural cooperation, akin to human-to-human interaction. Demonstrations included real-time speech translation, detecting animal mentions in stories, and monitoring posture. While the technology is not yet publicly available, Thinking Machines plans to launch a limited research preview in the coming months, with a wider release scheduled for later in 2026. The announcement comes as the company navigates staff changes, including departures to Meta and returns to OpenAI, following its establishment in February 2025.
The Verge