Huawei Cloud has officially launched the open-source Pangu model, openPangu-2.1-Pro, on its MaaS (Model as a Service) platform. This latest iteration of the Pangu model brings significant upgrades focused on three key areas: Agent development, AI coding, and inference optimization. These enhancements aim to bolster capabilities in crucial domains such as in-depth research, HarmonyOS system control, HarmonyOS App development, and Ascend operator generation and inference link optimization.
Enhanced Agent Capabilities
In the realm of Agent development, openPangu-2.1-Pro has been fine-tuned with over 100,000 specialized data points. This optimization specifically targets the skills triggering, tool selection, and invocation capabilities, leading to more effective complex task planning and execution. The model shows marked improvements in scenarios like in-depth research and controlling the HarmonyOS ecosystem.
Advancements in AI Coding
The new model places a strong emphasis on optimizing for HarmonyOS App development and Ascend operator generation. For HarmonyOS App development, it continues to refine the accuracy and applicability of generated code, particularly for the ArkTS language. Developers working with HarmonyOS will likely experience improved code generation that is more practical and easier to implement.
Furthermore, openPangu-2.1-Pro has undergone specialized optimization for the Ascend operator development workflow. This enhancement aims to boost the generation of professional-grade code for Ascend AI processors, thereby lowering the development threshold and increasing R&D efficiency for developers in this specialized field.
Optimized Inference and Deployment
Significant improvements have been made to the inference process through optimizations in key operators and the entire inference link. In typical deployment scenarios, these optimizations result in a substantial increase in throughput efficiency, allowing applications to handle high concurrency with more efficient resource utilization.
Huawei Cloud’s MaaS platform further facilitates the integration and operation of openPangu-2.1-Pro. Users can benefit from pre-built model APIs, rate limiting, usage statistics, and performance monitoring capabilities, enabling rapid deployment and effective management.
Integration with Core Huawei Products
Looking ahead, openPangu-2.1-Pro is set to be integrated into several core Huawei Cloud products, including the AI Agent development platform AgentArts, the AI coding assistant CodeArts, and the enterprise productivity suite OfficeAce. This integration is expected to accelerate the application of the model’s capabilities across various practical business scenarios, spanning intelligent agent development, AI-powered coding, and enterprise office automation.
Pricing Information
Huawei Cloud has also released the pricing details for openPangu-2.1-Pro:
- Inference Input: 0.8 Yuan / Million Tokens
- Inference Output: 3.2 Yuan / Million Tokens
- Cache Hit Input: 0.016 Yuan / Million Tokens
