Thinking Machines shows off preview of near-realtime AI voice and video conversation with new 'interaction models'

Advances state-of-the-art in real-time multimodal interaction, pushing AI capabilities forward.

Demonstrates innovation from a well-funded startup, encouraging investment in AI startups.

Creates new business opportunities for real-time AI applications and strengthens Thinking Machines' market position.

Announced breakthrough interaction models, strengthening its position as an AI innovator.

Former OpenAI CTO, co-founder of Thinking Machines, leading high-profile AI research.

Competitor whose GPT-realtime-2.0 lags in latency and interaction quality.

Gemini-3.1-flash-live has higher latency; competition may pressure Google.

Partner for deploying the models on Vera Rubin systems; benefits from increased AI compute demand.

Lead investor in Thinking Machines, likely to see returns if models succeed.

Attempted to acquire Thinking Machines and hired several employees, but also lost talent to them.

New attack surface for adversarial inputs in real-time streams.

Real-time processing of voice/video raises significant data handling and privacy concerns.

If models behave unsafely, it could damage company reputation.

Scaling from research preview to reliable product is challenging.

Cloud infrastructure is sufficient; no new risks identified.

No immediate geopolitical tension; but AI leadership competition may increase.

Real-time voice/video AI may face strict regulations on data privacy and consent.

Dependence on Nvidia hardware could be a bottleneck.

May replace some customer service jobs but also create new AI-related roles.

Real-time AI actions in physical domains could cause harm; liability unclear.