Evaluation surpasses Google and Nvidia! Alibaba releases RynnBrain robot model: enabling robots to possess a "thinking brain"

```

On February 10, Alibaba officially launched the RynnBrain foundational AI model for robotics. This open-source model aims to endow robots with perception, decision-making, and execution abilities, driving autonomous task completion in real-world scenarios.

RynnBrain is independently developed by Alibaba DAMO Academy and possesses core capabilities such as environmental interaction, spatiotemporal understanding, and task decomposition and planning. The model can help robots complete object recognition and localization, predict movement trajectories, and achieve precise navigation and autonomous operation in dynamically complex environments such as kitchens and factory production lines.

According to test data released by Alibaba, RynnBrain performed outstandingly in multiple authoritative evaluations, surpassing industry mainstream models such as Google Gemini Robotics-ER 1.5 and Nvidia Cosmos-Reason2. The model has set new records (SOTA) in 16 embodied open source evaluation leaderboards.

Currently, robotics technology is becoming a key track in global technological competition and industrial transformation, with cutting-edge directions such as humanoid robots seen as major drivers in reshaping the manufacturing and service industries. Alibaba’s release of a foundational model with a "thinking brain" not only reflects its continued investment in core AI technology, but also demonstrates a clear path for standardizing technology and driving real-world adoption.

Breakthrough in Spatiotemporal Memory and Reasoning Ability

RynnBrain's core technological breakthrough is the first integration of spatiotemporal memory and spatial reasoning abilities into robotic systems. By embedding these two key abilities in the model architecture, robots can maintain continuity and consistency of their working state when executing multiple tasks.

In practical applications, if a robot equipped with this model is interrupted while performing Task A and switches to Task B, it can accurately remember the spatiotemporal points and execution progress of Task A, and autonomously resume the previously interrupted workflow after Task B is completed.

The model integrates multidimensional abilities such as environmental cognition, precise localization, logical reasoning, and task planning, and shows strong scalability. Based on the RynnBrain foundational framework, developers only need hundreds of samples for fine-tuning to efficiently train dedicated models for different scenarios such as navigation, planning, and motion control.

Fully Open-Source Strategy

This time, DAMO Academy has open-sourced all seven models in the RynnBrain series, covering a variety of specifications from 2 billion parameter versions to a 30B Mixture-of-Experts (MoE) architecture. The series is trained based on the Qwen3-VL vision-language model and is now available on platforms such as Hugging Face and GitHub.

Among them, the industry's first 30B embodied model using the MoE architecture aims to improve robots’ motion fluency and response speed. To standardize evaluation measures, DAMO Academy has concurrently launched a new benchmark, RynnBrain-Bench, focused on spatiotemporal fine-grained task evaluation, filling the current gap in industry assessments for this domain.

Zhao Deli, head of the DAMO Academy Embodied Intelligence Lab, said:

"RynnBrain has, for the first time, enabled the brain to deeply understand and reliably plan for the physical world, taking a crucial step toward general embodied intelligence based on a hierarchical big-brain and small-brain architecture. We look forward to it accelerating AI's transition from the digital world into real physical scenarios."

Accelerating Industrialization of Embodied Intelligence

Chinese technology companies continue to ramp up open-source efforts in artificial intelligence, forming a technology development path characterized by open collaboration. In cutting-edge fields such as embodied intelligence, open-source strategies help aggregate global developer resources and accelerate technology iteration and ecosystem construction.

Robotics technology is seen as a key area for driving industrial upgrading. At the policy level, smart robots—including humanoid robots—have been clearly identified as priority development areas, with the aim of reshaping the operating models of manufacturing and services through technological innovation.

DAMO Academy continues to advance technology openness in this field. It has successively open-sourced several embodied intelligence models, including the WorldVLA (combining world models with vision-language models) and the environment understanding model RynnEC, and has released the industry’s first robot context protocol RynnRCP, aiming to build a deployable, scalable, and continuously evolving embodied intelligence system.

Risk Warning and DisclaimerThe market carries risk, and investments must be made with caution. This article does not constitute personal investment advice and does not take into account the individual investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific situation. Investments made accordingly are at your own risk. ```