Huawei Cloud to take openPangu-2.0-Flash and Pro offline from MaaS on November 30

Huawei Cloud will stop selling its openPangu-2.0-Flash and openPangu-2.0-Pro services on its MaaS (Model as a Service) platform at 00:00 Beijing time on November 6, and will take both offline at 00:00 on November 30, according to a model retirement notice on the Huawei Cloud support site. IT Home reported the notice on November 3, and Pandaily covered the report.

Retirement schedule

• November 6, 00:00 Beijing time: new users can no longer purchase or activate either service.
• Through the end of November: existing customers can keep calling the models.
• November 30, 00:00: neither new nor existing users can call either model.

The notice lists the services in Huawei Cloud's CN Southwest-Guiyang1 region.

Recommended replacements

Retiring service	Version	Recommended replacement
openPangu-2.0-Flash	20260630	DeepSeek-V4.1-Flash-1017
openPangu-2.0-Pro	20260730	Zhipu's GLM-5.3

Huawei Cloud says the models differ in capabilities, API parameters, pricing and real-world task performance. It asks customers to read the model documentation and run their own evaluations before switching.

The migration guide splits customers into two groups. Those on the MaaS standard API V1 need to move to the OpenAI-compatible endpoint. Those already on API V2 or the OpenAI-compatible interface only need to change the model parameter.

Next openPangu generation

According to IT Home, Huawei Cloud said a new generation of openPangu models is being planned and will be launched step by step. It has not disclosed model names or a timeline.

Background on the models

The two models were among Huawei's largest open releases this year. The 92-billion-parameter openPangu-2.0-Flash was open-sourced at the end of June. openPangu-2.0-Pro, a 505-billion-parameter sparse model trained on Ascend NPUs, followed in late July. In September, Huawei also released pretraining, supervised fine-tuning and reinforcement learning code for the series. The retirement notice covers only the hosted MaaS services and does not mention the open-source releases.

Other retirements in the same bulletin

Model turnover on the platform is frequent. The same bulletin schedules GLM-5.1 for retirement on November 16 with GLM-5.3 as its successor, moves DeepSeek-V4-Flash users to DeepSeek-V4.1-Flash, and retires Kimi-K2.6 on November 8. It also lists several Qwen and Wan models for retirement. Earlier notices retired DeepSeek-V4-Pro in September and several DeepSeek, Qwen and GLM versions over the summer.

What this means

For MaaS customers, workloads built on openPangu-2.0 must move to other hosted models on the platform, at least until the next openPangu generation becomes available. Callers on the standard API V1 face a larger change, since they must adopt the OpenAI-compatible endpoint, while API V2 and OpenAI-compatible users only change the model parameter. Because Huawei Cloud itself warns that capabilities, parameters and pricing differ, teams should test the recommended replacements against their own tasks before the November 30 cutoff. The notice affects only the hosted services, and the open-source openPangu-2.0 releases are not addressed in it.