Alibaba has officially launched RynnBrain, a foundational AI model for robotics, on February 10. This open-source model is designed to equip robots with perception, decision-making, and execution capabilities, enhancing their ability to complete tasks autonomously in real-world scenarios.
Developed independently by Alibaba's DAMO Academy, RynnBrain features core competencies such as environmental interaction, spatiotemporal understanding, and task decomposition and planning. The model assists robots in object recognition and localization, motion trajectory prediction, and enables precise navigation and autonomous operations in dynamic and complex environments like kitchens and factory assembly lines.
According to test data released by Alibaba, RynnBrain has demonstrated outstanding performance in multiple authoritative evaluations, surpassing leading industry models such as Google's Gemini Robotics-ER 1.5 and NVIDIA's Cosmos-Reason2. The model has set new records (SOTA) in 16 embodied AI open-source evaluation benchmarks.
Robotics technology is increasingly becoming a key area in global technological competition and industrial transformation, with advanced directions like humanoid robots regarded as major drivers reshaping the ecosystems of manufacturing and service industries. Alibaba's release of this foundational model, which possesses "thinking brain" attributes, not only reflects the company's continued investment in core AI technologies but also demonstrates a clear pathway toward promoting technical standardization and industrial application.
A breakthrough in RynnBrain's core technology lies in its integration of spatiotemporal memory and spatial reasoning capabilities into robotic systems for the first time. By embedding these key abilities into the model architecture, robots can maintain continuity and consistency in their working state when executing multiple tasks.
In practical applications, a robot equipped with this model can accurately remember the spatiotemporal context and progress of task A if interrupted to perform task B, and autonomously resume the previously interrupted workflow after completing task B.
The model combines multidimensional capabilities including environmental cognition, precise localization, logical reasoning, and task planning, showing strong scalability. Based on the RynnBrain framework, developers can efficiently train specialized models for various scenarios such as navigation, planning, and motion control by fine-tuning with just a few hundred data samples.
DAMO Academy has open-sourced all seven models in the RynnBrain series, covering specifications ranging from a 2-billion parameter version to a 30-billion parameter Mixture of Experts (MoE) architecture. The series, trained on the Qwen3-VL vision-language model, is now available on platforms like Hugging Face and GitHub.
Notably, the industry's first 30B embodied model using an MoE architecture aims to enhance the fluency and responsiveness of robotic movements. To standardize evaluation metrics, DAMO Academy also released a new benchmark called RynnBrain-Bench, focusing on fine-grained spatiotemporal task assessment, addressing a current gap in industry evaluation standards.
Zhao Deli, head of the Embodied Intelligence Laboratory at DAMO Academy, stated that RynnBrain represents a crucial step toward general embodied intelligence under a hierarchical brain architecture, enabling deep understanding and reliable planning of the physical world. The academy expects the model to accelerate the process of AI transitioning from the digital world to real physical scenarios.
Chinese tech companies are increasingly investing in open-source initiatives in the AI field, fostering a technology development path characterized by open collaboration. In cutting-edge areas like embodied intelligence, open-source strategies help pool global developer resources, accelerating technological iteration and ecosystem building.
Robotics technology is seen as a key driver of industrial upgrading. At the policy level, intelligent robots, including humanoid robots, have been designated as priority development directions, aiming to reshape operational models in manufacturing and services through technological innovation.
DAMO Academy continues to advance open technology in this field, having previously open-sourced several embodied intelligence models such as WorldVLA, which integrates world models with vision-language models, and the environmental comprehension model RynnEC. The academy has also released the industry's first robot context protocol, RynnRCP, committed to building deployable, scalable, and continuously evolvable embodied intelligence systems.