Huawei on June 30 said it open-sourced and launched openPangu-2.0-Flash, a 92-billion-parameter AI model.

The release adds to Huawei’s Pangu model family rather than starting a new one. Pangu was first launched in July 2021 and later expanded into industry-specific versions through Huawei Cloud.

Huawei described openPangu as its open-source AI model brand. The company said the brand is meant to provide references for Ascend-native training and inference technologies, support AI agents, and speed up commercial AI adoption and ecosystem development.

Huawei also said it delivers 2x single-card throughput on Ascend compared with other mainstream open-source models, although the materials reviewed did not include standard benchmark scores.

🔗 Source: Security Times

Recent Huawei developments