Zhiyuan AI

Zhiyuan AI refers to Beijing Zhiyuan Artificial Intelligence Research Institute (BAAI), established in 2018 with support from the Ministry of Science and Technology and the Beijing Municipal Government. It is a non-profit, new-type R&D institution positioned as a national-level flagship for open-source AI innovation, focused on frontier exploration and open-source ecosystem construction for foundational technologies such as large models.

Representative achievements include the 'Wudao' series of large models (e.g., Wudao 2.0 with 1.75 trillion parameters), the open-source Aquila series, BGE text embedding models, and multimodal models such as Emu and Bunny, all offering free commercial-use licenses.

FlagOpen is Zhiyuan Research Institute's full-stack open-source technology system for large models, including FlagAI (algorithm and model toolset), FlagPerf (hardware evaluation engine), and FlagEval (model evaluation tools), designed to lower development barriers and support multiple frameworks and chips.

You can use Zhiyuan Institute's 'Zhiyuan Index (CUGE)' Chinese evaluation benchmark and its FlagEval open-source evaluation framework. They provide multi-dimensional, scientifically fair evaluation tools and results, offering authoritative references for industry selection.

Zhiyuan Institute spearheaded the publication of the world's first 'Global AI Deception Systematic Report', first proposed a unified AI Deception Lifecycle Analysis Framework, and helped form the 'Beijing AI Safety International Consensus', warning about and researching the safety risks of cutting-edge AI systems.

By fully open-sourcing its models, tools, and evaluation systems, providing free commercial-use licenses; hosting Zhiyuan Conference and other academic activities; operating a developer community; and through the Zhiyuan Innovation Center promoting tech industry transformation to empower startups and enterprise growth.