Thinking Machines Launches Native Real-Time AI Model TML-276B
Thinking Machines — the AI startup founded by former OpenAI CTO Mira Murati — has launched TML-Interaction-Small 276B-A12B, a 276-billion-parameter model with 12 billion active parameters. Unlike existing voice and multimodal models that bolt real-time capability onto turn-based architectures, TML is trained from scratch for native real-time interaction. It eliminates the need for traditional voice activity detection (VAD). Observers describe it as leaving Google DeepMind and OpenAI behind in the real-time interaction space.
Why It Matters
The model reframes the AI bottleneck: FLOPS and intelligence are no longer the constraint — human-to-AI bandwidth is. Thinking Machines is positioning TML as the I/O layer for real-time human-AI collaboration across meetings, education, and training applications.