MiniCPM5-2B open-sourced, claims top open-model rank in agent benchmarks
On September 8, 2026, ModelBest (面壁智能) open-sourced MiniCPM5-2B, a 2-billion-parameter end-side language model, along with training recipes, a reinforcement learning framework called Meshy, the JustRL II policy, and some SFT data. The company reports it ranks first among open-source models on the Artificial Analysis Intelligence Index with a score of 23, and first on agent workflow benchmarks with a score of 20, outperforming larger models such as Gemma 4 12B and Qwen3.5 9B.
Coverage timeline
机器之心机器之心
编辑｜陈陈 过去几年，端侧大模型一直在不断突破能力边界。 最初，人们关注的是手机等设备能否运行一个基础语言模型；随后，长文本理解、数学推理、代码生成、多语言能力逐渐被压缩进更小的模型。如今，随着模型能力进一步提升，一个新的问题开始浮现：Agent 能力，是否也能从云端走向端侧？ 这并不是简单地把一个聊天模型放进设备，一个 Agent，需要理解目标、拆解任务、调用工具、根据反馈调整策略，并在多轮交互中持续推进任务。 随着端侧模型能力不断提升，过去属于云端的复杂工作流，开始具备在设备侧独立运行的可能。 面壁智能正是在这一方向上的代表性机构。 9 月 8 日， 面壁智能开源新一代端侧语言基础模型 MiniCPM5-2B，让端侧通用 Agent 能力开始具备雏形 。该模型将 工具调用、深度搜索、代码生成 等智能体核心能力引入端侧，为复杂任务在设备侧运行提供了新的可能。 另外，团队此次开放的不只是模型权重和算法，还包括部分 训练 Recipe（如 RL recipe）、自研强化学习框架 Meshy、基于奖励的强化学习策略 JustRL II，以及一批 SFT 数据 。这在当前大模型开源生态中并不常见。 端侧模型轻装上阵，Agent 能力正在成形 MiniCPM5-2B 轻量但不代表能力弱。 根据 Artificial Analysis Intelligence Index 榜单，MiniCPM5-2B 在开源模型类别中排名第一。该榜单综合了 9 项评测，覆盖知识推理、代码能力以及 Agent 任务执行等多个维度。结果显示， MiniCPM5-2B 以 23 分领先 Gemma 4 12B、Qwen3.5 9B 等参数规模更大的模型 。 在衡量模型在 Agent 工作流中的榜单中， MiniCPM5-2B 以 20 分排名第一 ，领先于 Granite 4.2 8B（9 分）、Qwen3.5 9B（7 分）等参数规模更大的模型。 MiniCPM5-2B 在推理效率上也展现出优势 ，完成单个任务平均仅需要约 21K output tokens，与同规模模型相比处于较低水平。 值得注意的是，MiniCPM5-2B 在 真实工作任务评测中取得接近人类基准的成绩 ，这一结果表明，端侧模型的能力边界正在从简单问答进一步延伸到真实任务处理。 MiniCPM5-2B 也展现出 较高的智能密度
Related stories
Alibaba open-sources Qwen3.8-27B, a 27B multimodal dense model
Alibaba's Qwen team released the open-weights Qwen3.8-27B, a 27B-parameter native multimodal dense model under the Apache 2.0 license. It supports 262K native context (extendable to 1M tokens) and outperforms Qwen3.7-Plus in coding and office tasks, with independent benchmarks showing it matching GPT-5.6 Luna on the Artificial Analysis Intelligence Index. The model surpassed 1 million downloads within days of release.
11 reports · 4 sources1421 points · 788 comments
Z.ai confirms Ox Alpha is GLM-5.3 Flash, releases open weights
Z.ai confirmed that the anonymous model 'Ox Alpha', which went viral on OpenRouter, is a new iteration of its GLM series, specifically GLM-5.3 Flash, and released its weights on Hugging Face under a new license. The model, a 320B-parameter natively multimodal model running on domestic chips, topped OpenRouter's leaderboard and ended DeepSeek's 56-day streak on OpenCode. The license requires companies with over $10B in revenue to pass a security review.
9 reports · 4 sources801 points · 281 comments
Z.ai debuts GLM-5.3 with post-training scaling gains, open weights in two weeks
Z.ai (Zhipu AI) announced GLM-5.3, an open-source model built on the same base as GLM-5.2 with substantially extended post-training, yielding a 50% gain on internal coding benchmarks and state-of-the-art results among open models on Terminal Bench 3.0 and Agents' Last Exam (CLI). The model matches Mythos 5 in white-box code review and vulnerability discovery, and scored 60 on the Artificial Analysis Intelligence Index, tying Kimi K3. Model weights will be released in two weeks after security hardening; API pricing is set at $1.40 per million input tokens and $4.40 per million output tokens.
9 reports · 6 sources1154 points · 573 comments
Anthropic risk report raises misalignment risk to low, details stronger internal Model 2
Anthropic's second Risk Report, dated August 2026 and totaling 186 pages, raises its misalignment risk estimate from very low to low and states it does not plan to release an internal model called Model 2. The report describes Model 2 as a Mythos-class model that is slightly stronger overall than Claude Mythos 5 in some areas, already used extensively internally for coding, data generation, and agent tasks, but lacking full external evaluation and public release plans.
2 reports · 2 sources
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe