Thinking Machines and interaction models

2026-05-11 · Source: All posts - seangoedecke.com RSS feed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, medium

Summary

Thinking Machines has released "Interaction Models," their first AI model, after a year of development and two billion dollars of capital. This system focuses on enhancing real-time conversational interaction rather than competing as a frontier model. It operates as a fully-duplex voice model, capable of simultaneous listening and speaking in 200ms "micro-turns," unlike traditional systems that alternate modes. A key innovation is its ability to delegate complex reasoning tasks to a separate, smarter language model in the background, integrating its output into the interaction model's context. Furthermore, Thinking Machines has scaled this model significantly, making it potentially twice the size of Moshi in active parameters and 40x in total parameters, and integrated video input to interpret facial expressions. This scale and multimodal capability represent a notable technical achievement, allowing for more human-like conversations.

Key takeaway

For AI Engineers developing conversational agents, if you are aiming for more natural, real-time human-AI interaction, consider adopting a fully-duplex architecture with delegated reasoning. Your systems can achieve lower latency and richer multimodal understanding by implementing micro-turns and offloading complex tasks to a separate, more powerful model. Explore optimizing inference for small, deterministic chunks and integrating video input to interpret user expressions, moving beyond traditional turn-based interactions.

Key insights

Scaled, multimodal fully-duplex models with delegated reasoning enhance real-time human-AI interaction beyond frontier models.

Principles

• Fully-duplex systems enable human-like interruptions.
• Delegate complex reasoning to a separate, smarter model.
• Multimodal input, like video, enhances conversational depth.

Method

The system uses 200ms micro-turns for simultaneous listening/speaking, delegating complex tasks to a background language model, and integrates video input. Inference libraries are optimized for tiny prefill/decode chunks.

In practice

• Implement micro-turns for low-latency interaction.
• Integrate external reasoning models via tool calls.
• Optimize inference for small, deterministic chunks.

Topics

• Interaction Models
• Fully-Duplex AI
• Multimodal Conversation
• Delegated Reasoning
• Real-time Inference
• Conversational AI

Code references

• kyutai-labs/moshi
• NVIDIA/personaplex

Best for: Research Scientist, AI Architect, NLP Engineer, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Thinking Machines drops a new, highly responsive model designed for humanlike interactions in real time — New AI architecture enables full-duplex, real-time multimodal interaction by processing data in 200-millisecond micro-turns.
• 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 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.
• 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.
• [AINews] Thinking Machines' Native Interaction Models - TML-Interaction-Small 276B-A12B - advances SOTA Realtime Voice and kills standard VAD — Real-time, full-duplex multimodal interaction models are advancing human-AI collaboration beyond turn-based systems.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by All posts - seangoedecke.com RSS feed.