The launch on the MaaS platform
Huawei Cloud has announced the arrival, on its MaaS (Model as a Service) platform, of the open-source model openPangu-2.1-Pro. The news, reported by IT之家 on November 15, therefore concerns the model's availability through a managed service: users can access it through preset APIs, with call throttling, usage metering, and performance monitoring provided by the platform itself.
According to the source, the update is not a simple version change but an intervention along three stated directions: agent development, code writing, and inference optimization. These axes correspond to areas in which the model is presented as enhanced, from deep research to control of the HarmonyOS operating system, to the generation of operators for Ascend.
Agents and code: what changes
On the agent front, openPangu-2.1-Pro has reportedly been optimized thanks to more than 100,000 specialized data items dedicated to skill activation and the selection and use of tools. The stated goal is to improve planning and execution of complex tasks, with effects on scenarios such as deep research and HarmonyOS control.
As for code, the model focuses on two applications: HarmonyOS app development and Ascend operator generation. In the first case, the ArkTS language is cited, with the aim of increasing the accuracy and usability of the produced code; in the second, there is mention of a specific adaptation to the operator development workflow, to reduce entry barriers and support programmers' efficiency. These are claims from the source, not independently measured results.
Performance, integrations, and pricing
Huawei Cloud claims that openPangu-2.1-Pro, thanks to the optimization of key operators and the inference path, achieves a substantial increase in throughput in typical deployment scenarios, with greater resource-use efficiency for high-concurrency applications. The figure is not accompanied by reference numbers or comparison parameters.
The model will also be integrated into cloud products named AgentArts, CodeArts, and OfficeAce, respectively for agent development, coding, and office. As for pricing, the source reports 0.8 yuan per million input tokens, 3.2 yuan per million output tokens, and 0.016 yuan per million tokens when a cache hit occurs.
The model's actual performance, the availability of the open-source model, and the methods of accessing the service remain to be independently verified; the material also does not clarify which components were actually opened and under what license.
