Agnes.AI国内官网升级上线 主打AI-Analyst功能进入综合排名前列
7月29日,人工智能平台Agnes.AI正式启用新官网agnes-ai.cn,并同步升级API访问地址。当天发布的2.5 Pro Alpha版本在综合智能表现排名进入前十。此次服务升级聚焦模型应用落地和开发者使用体验优化。
官网服务同步升级 区分Pro版与Flash版定位
面对AI生态扩展趋势,Agnes.AI此次更新着重提升国内访问效率和开发便利性。原国际站账户无需重新注册,国内用户访问需将接口地址由https://apihub.agnes-ai.com/v1改为https://apihub.agnes-ai.cn/v1,程序参数保持不变。
平台新增API控制台入口(https://platform.agnes-ai.cn),区分不同用户需求体验。其中2.5 Pro Alpha版本针对性加强商业场景支持能力,不限期开放基础功能的2.5 Flash版本则帮助开发者接入更加灵活。
2.5 Pro Alpha在综合能力测试中排名第十
配合官网升级,最新研发的2.5 Pro Alpha版本公布了第三方评测成绩。根据Artificial Analysis公布的最新排名,该版本的智能综合得分达到39分,在153个评测模型中位列第九。这一评估维度包括终端编程、多轮对话连续理解等10余项工作场景模拟指标。
“Agnes.AI数据中心入口迁移与技术服务同步推进,新官网页面增加分析对话自动化、多模态数据整合等专业功能模块,这些改版也体现在前台服务交互细节中”
进阶优化多模态分析能力 定位企业级智能分析师
与当前多数AI平台主要提供单向搜索回答不同,Pro版专注企业用户数据洞察场景,整合支付/人流/舆情三类实时数据进行预测性分析。新增的云笔记功能可对会议文本进行记忆强化存储,并结合文档库自动提炼关键信息,方便用户后续快速回溯重点。
新官网还开设"模型训练实验室",国内开发者可使用最新的2.5训练架构进行算法调试,调用白皮书代码样例参考训练过程。论坛板块采用类似GitHub的投票体系,优先采用得到较多开发者验证的模型版本。

Agnes-2-5 Pro Alpha Model Achieves High-Complexity Task Execution and 8.16 Trillion Weekly Token Processing
Agnes-2.5 series models showcase distinct functionalities for coding and development tasks, with Pro Alpha targeting advanced scenarios while Flash remains free for routine applications.
Platform Model Delineation
The Agnes-2.5 Flash model is designed for daily coding duties, supporting tasks like document handling, code generation, and bug fixing without any cost limitations. In contrast, the Agnes-2.5 Pro Alpha serves complex domains such as intricate system interactions and high-level software engineering.
- Flash emphasizes accessibility for common development roles, enabling users to perform tasks without financial barriers.
- Pro Alpha correlates with premium capabilities, including support for 1 million token context windows and analysis of large-scale projects.
Performance Metrics from Agnes Platform
Recent disclosures from Agnes indicate that the fully multimodal model processes 8.16 trillion tokens weekly, composing 5.10 trillion text tokens and 3.06 trillion multimodal tokens, with text constituting 62.5% of the total volume.
The platform demonstrates a shift toward multimodal applications, indicating increased token usage in image and video tasks, surpassing one-third of the weekly load despite text still being predominant.
Assessment Through Real-World Scenarios
Two distinct evaluations demonstrate the model's operational efficiency. The first task involved generating an interactive text-based game prototype, where the model independently defined elements like story themes and user interfaces.
The second case required building a color echo tool, enabling on-device processing for image manipulation without backend dependency.
- In the game development test, Agnes-2.5 Pro Alpha created a playable web prototype with features like branching narratives and visual elements.
- For the color echo application, the model integrated functions such as color transfer and export mechanisms through incremental refinement.
Ongoing Platform Activity and Developer Adoption
The Agnes system handles diverse tasks, with the performance data reflecting current usage patterns aligned with its infrastructure enhancements.
These developments reinforce the platform's utility for both individual creators and complex collaborative projects in the coding sphere.

【技术进展】Agnes团队为开源社区贡献四项优化,助力大模型推理服务运行效率提升
Agnes团队近日在其开源工作之外,持续参与SGLang社区建设。根据GitHub公开记录显示,团队提交的四项Pull Request已成功合并。相关工作的核心方向包括多卡并行、显存优化、缓存Token统计及KV Cache稳定性,这些都旨在提升大模型推理服务的运行效率与系统可靠性。
大模型推理服务效率提升关键技术
众所周知,显存优化是大模型推理过程中的关键技术,而缓存Token统计则是为了保证模型在长时间运行中的稳定性。KV Cache是大模型推理过程中的关键数据结构,用于存储中间计算结果。
2.5 Pro Alpha版本测试进展
随着公司最新一代大模型Agnes 2.5 Pro Alpha版本的推出,性能优化工作正在稳步推进。当前,这个开发者实测阶段的版本已经可以通过官网API接口进行体验。正式版本预计即将发布,为用户提供全面提升的推理服务体验。
- 多卡并行技术:实现GPU资源的最大化利用
- 显存优化方案:降低模型推理过程中的内存占用
- 缓存Token统计机制:提升计算过程的准确性
- KV Cache稳定性增强:保障长时间运行可靠性
本文信息源自新智元,官方网站提供进一步技术细节:https://agnes-ai.cn/
