ChainThink News: On June 30, according to official announcements, Huawei has officially open-sourced the PanGu openPangu-2.0-Flash model and will launch it on the Ascend Tribe community on GitCode.

The model features a 92B parameter MoE architecture with 6B active parameters and supports ultra-long context of up to 512K tokens. As per the open-source roadmap, initial release content includes model weights, foundational inference code, and training/inference operators.

Higher-performance openPangu-2.0-Pro model weights and inference code are scheduled for release in July, with additional components to be open-sourced progressively throughout the second half of the year.