Skip to content

Event

BAAI Releases FlagOS 2.2: Operators Increased to 1,454, Cross-Chip Performance Up 27%

1 reports · 1 sources

The Beijing Academy of Artificial Intelligence (BAAI) released FlagOS 2.2 on September 30 and presented its progress on October 17 at the 2026 AI Open Computing Conference and Zhongzhi FlagOS Technology Conference held in Beijing. In this version, the number of public operators in the nine major operator libraries increased from 564 to 1,454 (890 newly added); in tests covering over a thousand operators and 10 heterogeneous AI chips, cross-chip operator performance improved by about 27% compared with the previous version. At the same time, the FlagOS-Compressor weight conversion and quantization tool was released for the first time, and BAAI says a new model can be adapted to as many as 12 chips within one day. BAAI also presented joint optimization results for its embodied world model: after joint optimization of its new-generation embodied world model system with FlagOS's FlagGems operators, training framework, and distributed strategies, the VLM, Tx2V, and Action modules achieved speedups of approximately 2.3×, 1.9×, and 2.0×, respectively, with MFU at approximately 33%, 36%, and 35%, respectively. BAAI states that these figures are provided only as reference ranges and, because models and training settings differ, cannot be used to derive an overall performance ranking.

Report timeline

  1. BAAI's FlagOS 2.2 released: over 1,454 operators, 27% cross-chip performance gain

    BAAI presented FlagOS 2.2 progress at the 2026 AI Open Computing Conference and FlagOS Technical Conference; the version was released on September 30. Public operators in the nine operator libraries rose from 564 to 1,454 (890 added). In tests covering over a thousand operators and 10 heterogeneous AI chips, cross-chip operator performance improved by about 27% over the previous version. The FlagOS-Compressor weight conversion and quantization tool was released for the first time, enabling a new model to be adapted to as many as 12 chips within one day.

    智源研究院 · 微信公众号 ·

Related topics