Huawei Cloud will stop selling its two openPangu-2.0 model services on 6 November and take both offline on 30 November, Pandaily reported on 3 November, citing a retirement notice on Huawei Cloud’s support site. The decision covers openPangu-2.0-Flash and openPangu-2.0-Pro in the CN Southwest-Guiyang1 region of its Model-as-a-Service, or MaaS, platform, where customers access hosted models through APIs.
Key points
- New customers lose access to sales and activation of the two services on 6 November. Existing users can keep calling them until 30 November.
- Huawei Cloud recommends DeepSeek-V4.1-Flash-1017 for Flash users and Zhipu’s GLM-5.3 for Pro users.
- The migration guidance differs by API version, and Huawei Cloud asks customers to evaluate the replacement models before switching.
Huawei Cloud sets two cut-off dates
Sales and activation of openPangu-2.0-Flash and openPangu-2.0-Pro end at 00:00 Beijing time on 6 November. Customers already using either service can continue to make calls until the models are taken offline at 00:00 on 30 November. From that point, the hosted services will be unavailable to new and existing users.
Huawei Cloud’s recommended replacement for openPangu-2.0-Flash version 20260630 is DeepSeek-V4.1-Flash-1017. For openPangu-2.0-Pro version 20260730, it recommends GLM-5.3 from Zhipu. Both recommendations move users of Huawei’s own models to services built by other companies and hosted on its platform.
Huawei Cloud says the alternatives differ in capabilities, API parameters, prices and performance on actual tasks. It asks customers to consult the model documentation and conduct their own evaluations before migrating. That puts the choice of replacement model, and the assessment of how it performs in an existing workload, with the customer.
The changes required also depend on the interface a customer uses. Huawei Cloud’s migration guidance directs users of the MaaS standard API V1 to an OpenAI-compatible endpoint. Users already on API V2 or the OpenAI-compatible interface need to change the model parameter instead. Those are distinct migration paths for workloads using the same retiring services.
OpenPangu releases remain separate from MaaS
The retirement concerns hosted access to the models on MaaS. Huawei open-sourced openPangu-2.0-Flash at the end of June and openPangu-2.0-Pro in late July. In September, it also released code for the series covering pretraining, supervised fine-tuning and reinforcement learning. The MaaS notice does not address those open-source releases.
Pandaily reported 92 billion parameters for Flash and 505 billion for Pro, describing the latter as a sparse model trained on Ascend NPUs. A MarsBit report published by KuCoin gives 9.2 billion and 50.5 billion respectively in its summary. Its main text gives 92 billion and 505 billion.
MarsBit dated Huawei Cloud’s announcement to 1 November. IT Home reported the retirement notice on 3 November, Pandaily reported. The two accounts give different dates for those steps.
Huawei Cloud says it is planning another generation of openPangu models for release in stages. The company has not given model names or a timetable. Customers using the hosted Flash and Pro services therefore have the two named third-party recommendations for the scheduled migration.
GLM-5.1 and Kimi-K2.6 also face retirement
The bulletin covers other services on Huawei Cloud’s platform. It schedules Kimi-K2.6 for retirement on 8 November and directs users of DeepSeek-V4-Flash to DeepSeek-V4.1-Flash. It also lists several Qwen and Wan models for retirement. Earlier platform notices retired DeepSeek-V4-Pro in September and versions of DeepSeek, Qwen and GLM over the summer.
Huawei Cloud will retire GLM-5.1 on 16 November and direct its users to GLM-5.3.
