Thinking Machines

Thinking Machines released TML-Interaction-Small 276B-A12B, a 276-billion-parameter MoE model with only 12 billion active parameters, achieving sub-200ms end-to-end latency, surpassing GPT-4o and Gemini 3.1-Flash in real-time voice interaction, time-aligned microturns, and visual proactivity, effectively eliminating standard VAD.

Interaction models enable native real-time multimodal interaction, overcoming the limitations of traditional turn-based interfaces and significantly enhancing human-AI collaboration efficiency.

This episode highlights OpenAI's real-time voice API, Thinking Machines' interactive model architecture, Anthropic's legal product, and market dynamics, revealing advancements in real-time interaction, vertical applications, and security compliance.

Thinking Machines' latest launch approach is impressive, showcasing through platform-native content, which is more effective than traditional methods.

Still early days, but the naturalness of interactions with multimodal AI models is rapidly improving; real-time, low-latency capabilities are unlocking applications unimaginable today, reshaping human-AI collaboration.

Thinking Machines introduces a native multimodal Interaction Model enabling 200ms-level real-time collaboration, breaking traditional Agent framework limitations by integrating all modalities into a single architecture. The model supports instant user interruptions through frontend-backend model coordination, significantly improving conversational naturalness.

The article discusses advances in real-time multimodal interaction models, emphasizing their closeness to real human interaction.

Lilian Weng advocates for increased collaboration in human-AI interactivity and announces a $100,000 grant program by Thinking Machines, but the post lacks technical depth and serves primarily as a promotional announcement.

Human-to-human collaboration plays a key role in improving human-AI collaboration, as learned from maintaining a 137-page training log across 12 iterations.