FlagOS 2.2 expands cross-chip AI optimization
智源众智 released FlagOS 2.2 on September 30, 2026, and presented it at the AI Open Computing Conference in Beijing on October 17. The release adds nine operator libraries covering 1,454 public operators, including FlagGems-SGLang, FlagOS-Compressor, KernelGen and FlagPrism; vendor-reported tests showed about 27% higher cross-chip operator performance across more than 1,000 operators and 10 heterogeneous AI chips. In a GLM-5.3-Flash deployment, the optimized Ascend 910B service reached 2.04 times the prior logical throughput, while a scenario-based rental calculation placed its compute cost 19.4% to 27.6% below the corresponding commercial API pricing reference.
FlagOS 2.2 targets a practical infrastructure problem: adapting rapidly changing models to heterogeneous chips and long-lived compute resources through shared operators, compilers, quantization and deployment tools rather than separate adaptation work for every hardware target.
2026 年 10 月 17 日，以“一栈贯通，智算无界”为主题的 2026 人工智能开放计算大会暨众智 FlagOS 技术大会在北京举行；继 9 月 30 日 FlagOS 2.2 发布后，大会展示了从底层算子、编译器到模型部署与服务的技术进展，并提出 AI 模型快速换代而算力基础设施需长期服役的现实问题。
智源众智 FlagOS 2.2 从多芯适配走向跨芯复用，九大算子库合计覆盖 1454 个算子、较上一版增加 890 个，在上千个算子、10 款异构 AI 芯片测试范围内跨芯算子性能提升约 27%；GLM-5.3-Flash 实践中，存量昇腾 910B 逻辑服务吞吐达优化前 2.04 倍，指定条件下按租金折算的算力成本较对应商业 API 计费参考低 19.4%—27.6%，世界模型和 ARM 端侧推理也出现可量化的联合优化成果。
#FlagOS 2.2#FlagGems-SGLang#FlagOS-Compressor#KernelGen#FlagPrism#GLM-5.3-Flash#Ascend 910B