Thinking Machines Lab has launched ‘interaction models,’ a novel AI system designed for fluid, real-time conversations by processing and responding continuously, moving away from traditional delayed-response models.
- AI responds continuously during conversations rather than waiting for prompts
- Supports simultaneous speaking, time awareness, and contextual interjections
- Currently in research preview with plans for wider release in 2026
What happened
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, introduced a new form of AI called ‘interaction models’ that allows for real-time, dynamic conversations. Unlike most existing AI systems which wait for a user to finish typing or speaking before responding, these models can process and respond simultaneously, enabling a more natural flow of communication.
The system, called TML-Interaction-Small, is currently in a research preview phase and not yet publicly accessible. Key capabilities include managing dialogue seamlessly by tracking speaker intent, handling verbal and visual interjections mid-conversation, and allowing both AI and user to speak at once. The lab plans a limited preview release soon, followed by broader access later in the year.
Why it matters
By enabling continuous interaction, this technology aims to shift AI from mere automated responders to collaborative partners that keep humans actively engaged. This approach contrasts with today’s prevalent AI design, which often pauses conversation until a full input is received, thereby slowing collaboration and limiting usefulness in real-world tasks.
The interactive model’s design fosters constant human oversight and incremental feedback during tasks, addressing current AI limitations that lack awareness of users’ ongoing actions. If successful, it represents a significant step toward fluid, human-like AI conversations that could improve applications ranging from customer service to real-time translation.
What to watch next
While promising, current versions of the interaction models face challenges with lengthy conversations due to context overflow and require robust internet connectivity for smooth operation. Additionally, larger-scale models are not yet capable of real-time responsiveness, highlighting the engineering work needed to scale the technology effectively.
The AI community is watching closely as the lab prepares limited public trials later this year. The real test will be how well these models perform in everyday, diverse scenarios outside controlled research environments and whether they can indeed redefine human-AI interaction by keeping users continuously engaged and collaborating.