Article · AI translation
IT Home reported on November 15 that Huawei Cloud announced today the official launch of the open-source Pangu model openPangu-2.1-Pro on its MaaS model-as-a-service platform.
The model has been upgraded in 3 areas: Agent development, AI coding, and inference optimization. It offers enhanced capabilities in scenarios including deep research, HarmonyOS operation, HarmonyOS app development, Ascend operator generation, and inference pipeline optimization.
For Agent capabilities, openPangu-2.1-Pro uses more than 100,000 specialized data entries to optimize Skills triggering, tool selection, and tool invocation, improving the planning and execution of complex tasks and enhancing its capabilities in scenarios such as deep research and HarmonyOS operation.
For coding capabilities, the new model has been optimized and adapted specifically for HarmonyOS app development and Ascend operator generation. It continues to improve code generation accuracy and application usability for the HarmonyOS ArkTS language and related development tasks. Targeted optimizations for the Ascend operator development process strengthen its ability to generate specialized code, helping developers lower the barrier to development and improve R&D efficiency.
Through optimization of key operators and the inference pipeline, openPangu-2.1-Pro substantially increases throughput efficiency in typical deployment scenarios, supporting high-concurrency applications with more efficient resource use. Huawei Cloud MaaS also provides preset model APIs, rate limiting, usage statistics, and performance monitoring, enabling users to integrate and operate the model quickly.
openPangu-2.1-Pro will also be integrated into core Huawei Cloud products including Zhiguo AgentArts, M adao CodeArts, and Guoban OfficeAce, covering agent development, AI coding, and enterprise office work, and accelerating the extension of the model's capabilities into real business scenarios.
IT Home provides the following pricing:
Inference input: 0.8 yuan per 1 million tokens
Inference output: 3.2 yuan per 1 million tokens
Cache-hit input: 0.016 yuan per 1 million tokens