Woofun AI reports that Huawei has officially open-sourced the openPangu-2.0-Flash model, scheduled for release on the Ascend Tribe community via GitCode. The 92 billion parameter model utilizes a Mixture of Experts architecture, activating just 6 billion parameters while supporting a 512K context window. The initial package contains model weights, basic inference code, and training operators, with the higher-performance openPangu-2.0-Pro weights and code slated for July release.