Overview
- Alibaba’s DAMO Academy launched RynnBrain as an embodied-AI model for robotics, publishing multiple versions on GitHub and Hugging Face.
- Technical highlights include a 30‑billion‑parameter mixture‑of‑experts model that activates about 3 billion parameters at inference, spatiotemporal memory for perception, and a new RynnBrain‑Bench evaluation suite.
- Alibaba says RynnBrain set new highs across 16 embodied‑AI benchmarks and outperformed Google’s Gemini Robotics‑ER 1.5 and NVIDIA’s Cosmos‑Reason2, with results reported by the company and pending independent validation.
- Demonstrations show robots mapping objects, predicting trajectories, and executing step‑by‑step tasks, including a video of fruit being identified and sorted into a basket.
- RynnBrain is being integrated into Alibaba’s Physical AI stack alongside partnerships with NVIDIA and Joyson, with the push tied to cloud expansion plans and a modest stock uptick of about 0.30%.