华为云于 11 月 15 日在 MaaS 模型即服务平台正式上架开源盘古模型 openPangu-2.1-Pro 1。该版本重点针对 Agent 开发、AI 编码及推理优化三大方向进行了能力升级 1。
在具体技术细节上,openPangu-2.1-Pro 基于 10 万余条专项数据对 Agent 能力进行了优化 1。其代码生成能力特别适配了鸿蒙 ArkTS 语言以及昇腾算子的开发流程,增强了鸿蒙系统操控、鸿蒙 App 开发及昇腾算子生成等应用场景的表现 1。此外,模型还通过优化推理链路以支持深度研究任务 1。
目前,该模型已接入华为云智果 AgentArts、码道 CodeArts 和果办 OfficeAce 等多项核心产品 1。在 API 调用价格方面,推理输入成本为每百万 tokens 0.8 元,推理输出为每百万 tokens 3.2 元;若命中缓存,输入成本则降至每百万 tokens 0.016 元 1。
Huawei Cloud announced on November 15 that it has made its open-source model, openPangu-2.1-Pro, available on the MaaS (Model as a Service) platform 1. The release features significant upgrades in three key areas: agent development, AI coding, and inference optimization 1. To accelerate business scenario implementation, the model is integrated into several core Huawei Cloud products, including ZhiGuo AgentArts, MaDao CodeArts, and GuoBan OfficeAce 1.
The enhanced capabilities of openPangu-2.1-Pro support deep research, HarmonyOS system control, HarmonyOS app development, Ascend operator generation, and inference chain optimization 1. Specifically, the agent development functionality was trained using over 100,000 specialized data entries 1. Regarding coding capabilities, the model is adapted to focus on the HarmonyOS ArkTS language and the workflow for developing Ascend operators 1.
Huawei Cloud also disclosed API pricing details based on token usage 1. The price for inference input is 0.8 yuan per million tokens, while inference output costs 3.2 yuan per million tokens 1. Additionally, cached input hits are priced at 0.016 yuan per million tokens 1.
评论
还没有评论,欢迎留下第一条。