Huawei releases open-source large-scale model Pangu 2.0: up to 505 billion parameters and 512K context.

CoinFeed reported on June 12 that Huawei released openPangu 2.0, an open-source large-scale model, at HDC 2026. It offers two versions: Pro and Flash, both supporting 512K context. The 2.0 Pro version has a total of 505 billion parameters and 18 billion activation parameters, while the 2.0 Flash version has 92 billion parameters and 6 billion activation parameters. The official claim is that the model achieves twice the single-card throughput of other mainstream open-source models on Ascend computing power and is better adapted to HarmonyOS, improving the speed and accuracy of Agent tasks. Starting June 30, openPangu 2.0 plans to gradually open-source seven major components, including newly added open-source pre-training code, post-training code, and training operators.