Alibaba's RynnBrain Outperforms Nvidia's Cosmos-Reason2 on Robotics Benchmarks
Alibaba unveiled RynnBrain, an open-source AI foundation model trained on Qwen3-VL that maps objects, predicts trajectories and plans multi-step tasks in complex environments like factory lines and kitchens. Benchmark tests claim it outperforms Nvidia's Cosmos-Reason2 and Google's Gemini Robotics-ER 1.5 on standard robotics tasks.
1. RynnBrain's Technical Features
Alibaba’s DAMO Academy released RynnBrain as an open-source foundation model enabling robots to interact dynamically with real-world environments. It leverages the Qwen3-VL vision-language backbone to map object positions, predict motion trajectories and automatically sequence the steps needed to complete tasks in settings such as kitchens or assembly lines.
2. Benchmark Claims Versus Nvidia
The company reports that RynnBrain surpassed state-of-the-art scores when pitted against Nvidia’s proprietary Cosmos-Reason2 and Google’s Gemini Robotics-ER 1.5 on multiple standard robotics evaluation suites. These benchmarks measure accuracy in object manipulation, trajectory planning and environment navigation.
3. Implications for Nvidia’s AI Strategy
By offering cutting-edge performance under an open-source license, RynnBrain could pressure Nvidia to adjust its product positioning or accelerate feature releases for Cosmos-Reason2. The shift toward open-source robotics models may erode some competitive advantages Nvidia holds in proprietary AI tooling.