On July 21, 2026, the GOAI Global Open-Source AI Challenge (goaihz.com) officially unveiled its four competition tracks and participation rules, calling for projects from AI builders, research teams, open-source communities, and industry innovators worldwide.

The GOAI Global Open-source AI Challenge is hosted by the Hangzhou Open Source Artificial Intelligence Foundation, with the Agentic AI Foundation (AAIF) and the LF AI & Data Foundation serving as Global Open-source Partners. The Challenge is undertaken by Zhejiang Lab, Alibaba Cloud Intelligence Group, Ant Group Co., Ltd., Hangzhou AILume Future Technology Co., Ltd., Yunshenchu Technology Co., Ltd., Manycore Tech Inc., and other leading organizations.

Four Tracks, Four Pathways to Real-World AI Implementation

Track 1: Agent Infra

The Agent Infra track focuses on multi-agent infrastructure and collaborative systems for complex tasks, driving agents from the Demo phase to Production.

Participating teams must start from real-world scenarios and, using AgentTeams as the foundation for collaborative design, build an end-to-end closed-loop task system comprising at least three Agents with different roles. They must fully demonstrate capabilities such as task decomposition, context passing, tool invocation, result verification, and the documentation of execution evidence, and encapsulate the key capabilities into reusable Skills.

This track places greater emphasis on system engineering capabilities—specifically, the stability of multi-agent collaboration, the level of engineering sophistication of skills, and reliable integration with enterprise systems.

Track 2: Boundless Agents

The Boundless Agents track focuses on real-world industry scenarios and is organized into five major categories: “AI + Smart Glasses,” “AI + Automotive,” “AI + Finance,” “AI + Education,” and “AI + Industrial Manufacturing.” It encourages participating teams to develop functional, demonstrable, and replicable agent applications.

Entries must form a clear, closed-loop task process covering user input, task understanding, workflow orchestration, knowledge base or tool invocation, result delivery, exception handling and effectiveness verification. They must also specify the target users, pain points in the scenario, data sources, privacy protection measures, and the boundaries of industry application.

This track seeks more than just an attractive demonstration of features; it seeks application prototypes that can truly integrate into users’ workflows, improve product experiences, or enhance business efficiency. The core challenge of this track is how to enable intelligent agents to consistently create value in real-world scenarios.

Track 3: AI for Research

The Frontier Exploration track focuses on the interdisciplinary applications of AI and scientific research, offering two distinct competition pathways: algorithm challenges and open exploration challenges.

Algorithm challenges cover areas such as virtual cells, small-molecule–protein binding trajectory prediction, and literature-driven scientific discovery agents in materials science. Participating teams must develop a working, reproducible algorithmic solution based on the provided dataset and evaluation framework, and explain the scientific significance of their results.

The Open Exploration track encourages participants to draw from their own research fields to propose a real-world scientific problem for which there is no universally accepted answer, and to transform it into an environment where an Agent can enter, experiment, refine, and continuously explore.

This track seeks not only teams that “know how to use AI to conduct experiments,” but researchers capable of translating scientific intuition into computable problems, verifiable signals, and sustainable exploration environments.

Track 4: Embodied Future

The Embodied Future track focuses on the interaction between open embodied intelligence and the physical world, featuring two types of tasks: a general-purpose dual-arm collaborative manipulation test and an all-terrain patrol challenge in an industrial park.

The dual-arm collaborative manipulation test leverages the X-Eval simulation and real-world evaluation platform. It targets areas such as vision-language-action (VLA) models, world action models (WAM), dual-arm cooperative control, and simulation-to-reality transfer, assessing the system’s success rate, generalization ability, execution efficiency, and robustness in complex manipulation tasks.

The Industrial Park All-Terrain Patrol Challenge uses Lynx S10 of Yunshenchu Technology Co --- wheeled-legged robot as the base platform. It tests the robot’s multimodal perception, path planning, autonomous navigation, real-time obstacle avoidance, and task execution capabilities in complex environments such as slopes, gravel stairs, slippery surfaces, unknown obstacles, and dynamic disturbances.

Participating teams must submit a functional, reproducible, and measurable embodied intelligence solution. Through code, deployment instructions, data sources, technical documentation, and task validation results, they must demonstrate a complete closed-loop connection between model capabilities, systems engineering, and real-world task execution.

Schedule and Awards

The deadline for submitting preliminary entries for the “Track 1 Agent Infra,” “Track 2 Boundless Agents,” and “Track 3 AI for Research” is August 16; the deadline for the “Track 4 Embodied Future” is August 20. The finals and on-site demonstrations are scheduled to take place on September 22–23. Specific arrangements are subject to the organizing committee’s final notice.

The competition features one GOAI Grand Prize, with a cash prize of 1 million RMB. Each of the four tracks will also present its own track champion, first runner-up, second runner-up, and multiple special awards. Specific details regarding quotas, prize amounts, and selection criteria are subject to the participant handbook for each track and announcements on the official competition website.

Participants may register individually or as a team of up to three members.

Visit the competition’s official website now at goaihz.com and click “Register Now” in the upper-right corner to participate. You can also download the competition handbooks for each track from the official website to learn more about the specific challenges.

Company：GOAI Global Open-source Al Challenge
Contact Person: Ada Wang
Email: goai@goaihz.com
Website: https://www.goaihz.com/
Telephone: 13522981265
City：Hangzhou, Zhejiang, China