面壁智能联合OpenBMB发布端侧模型MiniCPM5-2B,以2B参数量在AA-Index榜单取得4B以下模型最高分17分,平均分54.26,超越Qwen3.5-2B等竞品。
模型原生支持混合思考与512K上下文,已完成华为昇腾、英伟达等9款芯片的Day0适配,即将开源。
Mianbi Intelligence, in collaboration with OpenBMB, released the device-side model MiniCPM5-2B, achieving the highest score of 17 points for models below 4...
Mianbi Intelligence, in collaboration with OpenBMB, released the device-side model MiniCPM5-2B,
achieving the highest score of 17 points for models below 4B on the AA-Index with 2B parameters, with an average score of 54.26, surpassing competitors like Qwen 3.5-2B. The model natively supports hybrid thinking and 512K context, has completed Day 0 adaptation for nine chips including Huawei Ascend and Nvidia, and is about to be open source.
面壁智能联合OpenBMB发布端侧模型MiniCPM5-2B,以2B参数量在AA-Index榜单取得4B以下模型最高分17分,平均分54.26,超越Qwen3.5-2B等竞品。
模型原生支持混合思考与512K上下文,已完成华为昇腾、英伟达等9款芯片的Day0适配,即将开源。
面壁智能联合OpenBMB发布端侧模型MiniCPM5-2B。公开材料称,该模型参数量为2B,在AA-Index榜单获17分、平均分54.26,并已完成9款芯片的Day0适配。
此次发布面向端侧模型场景。材料称,MiniCPM5-2B在4B以下模型中取得AA-Index最高分,并与Qwen3.5-2B等竞品进行比较;上述信息均来自发布方公开内容。
Aioga判断,2B参数规模、512K上下文、混合思考与多芯片适配,是此次更新值得关注的组合;但现有材料未披露测试细节与第三方复核结果,暂不宜扩大性能结论。
Aioga判断,若榜单表现和芯片适配情况在开源后得到复现,该模型可能为端侧部署提供新的候选方案。其实际速度、资源占用、兼容性与场景效果仍需后续验证。 值得关注模型按计划开源后的权重、代码与许可信息,并核验AA-Index测试设置、各项得分及9款芯片适配范围;同时应通过独立测试确认512K上下文和混合思考的实际表现。
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Ingestion channel: Summary aggregation · Source domain: mp.weixin.qq.com
Source: 公众号:面壁智能(MiniCPM)
Original link: Open original source
Aioga archive: Open intelligence page
Content record: summary-fallback · Updated: 2026-07-19T13:12:39.000Z

统一接入主流 AI 模型 API,为开发、测试与生产环境提供稳定调用入口。
立即访问 api.w173.comAioga aggregates global AI updates and preserves source information for verification and citation.