Thinking Machines drops a new, highly responsive model designed for humanlike interactions in real time

2026-05-11 · Source: AI – SiliconANGLE · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

Thinking Machines Lab Inc., an AI research startup, has unveiled a research preview of its "interaction models," a new class of multimodal AI systems designed for real-time, humanlike interactions. These models overcome the typical pauses in AI communication by enabling "full-duplex" communication, allowing the AI to listen, see, and talk simultaneously. The core is a new architecture that processes inputs and outputs in 200-millisecond chunks, reacting instantly to visual or auditory cues. This "dual-model" architecture pairs TML-Interaction-Small, a 276-billion parameter mixture-of-experts model for rapid dialogue, with an asynchronous background agent for complex reasoning and web searches. This system achieved a turn-taking latency of less than 0.4 seconds on the FD-bench benchmark, outperforming Google's Gemini-3.1-flash-live (0.57 seconds) and GPT-realtime-2.0 (1.18 seconds).

Key takeaway

For CTOs and VPs of Engineering evaluating AI integration for high-stakes or customer-facing applications, Thinking Machines' interaction models offer a path to truly real-time, human-like AI. You should investigate this full-duplex architecture to reduce interaction latency and enable more natural, collaborative AI experiences, especially where immediate multimodal responsiveness is critical for operational efficiency or safety.

Key insights

New AI architecture enables full-duplex, real-time multimodal interaction by processing data in 200-millisecond micro-turns.

Principles

• AI must adapt to human interaction patterns.
• Low latency is critical for human-like collaboration.
• Dual-model architecture balances speed and deep reasoning.

Method

The system uses a multistream micro-turn-based design, processing inputs/outputs in 200ms chunks. It employs "encoder-free early fusion" for raw signal input directly into a lightweight embedding layer within the transformer.

In practice

• Monitor video feeds for real-time safety alerts.
• Enhance customer service calls with natural conversation flow.
• Manage time-sensitive requests without explicit timestamps.

Topics

• Interaction Models
• Full-Duplex AI
• Dual-Model Architecture
• TML-Interaction-Small
• Real-time AI

Best for: Machine Learning Engineer, CTO, VP of Engineering/Data, AI Engineer, AI Product Manager, Director of AI/ML

Related on AIssential

• Thinking Machines shows off preview of near-realtime AI voice and video conversation with new 'interaction models' — Thinking Machines' interaction models enable real-time, full-duplex human-AI communication by processing multimodal inputs and outputs simultaneously.
• Thinking Machines unveils AI that can interrupt users in real-time — Thinking Machines Lab developed a full-duplex AI model enabling real-time, interruptible human-AI conversation.
• Mira Murati’s Thinking Machines Lab Introduces Interaction Models: A Native Multimodal Architecture for Real-Time Human-AI Collaboration — Native multimodal architectures can enable true real-time human-AI collaboration by integrating interactivity into model weights.
• New Trends in AI Interaction Models — Thinking Machines Lab's Interaction Models enable full-duplex AI communication, allowing simultaneous listening and talking to eliminate conversational latency.
• Thinking Machines Just Announced More Human Like AI — Thinking Machines Lab introduces "Interaction Models" for real-time, human-like AI collaboration via simultaneous input/output.
• Interaction Models: A Scalable Approach to Human-AI Collaboration — Native, real-time multimodal interaction within AI models enhances human-AI collaboration beyond turn-based systems.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI – SiliconANGLE.