[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-agentflow-7b-icrl-oral-flow-grpo-14-9pct":3,"news-related-69e52a42-19c7-4580-8c49-5446233fbdde":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"69e52a42-19c7-4580-8c49-5446233fbdde","7B模型如何超越GPT-4o？ICLR Oral论文揭示AgentFlow流式训练新范式","当行业还在靠更大即更强堆参数时，一篇刚刚亮相 ICLR 2026 口头报告（Oral）的论文给出了不同的答案。来自多机构研究团队发表的 In-the-Flow Agentic System Optimization for Effective Planning and Tool Use，提出了一个名为 AgentFlow 的新型智能体架构——仅用 7B 参数的基础模型，在搜索、数学、科学推理等十项基准上全面超越了 GPT-4o，且平均领先幅度达 14.9%。\n\n这背后的核心创新并非更大的模型，而是一套全新的模块化智能体框架与在线强化学习训练方法。\n\n当前主流的 Tool-Augmented LLM（工具增强型语言模型）普遍采用单一整体策略，将思考链与工具调用混合在完整上下文中。这种单体式设计在任务链较短、工具种类有限时尚可运作，但一旦面临长周期规划或多样化工具调用，上下文的线性增长便导致效率急剧下降，且泛化能力薄弱。\n\nAgentFlow 则将智能体拆解为四个各司其职的模块：Planner（规划器）负责理解任务并拆解步骤；Executor（执行器）调用工具并返回结果；Verifier（验证器）核查中间输出的正确性；Generator（生成器）负责记忆管理与上下文构建。四模块通过进化记忆动态协调，每个模块各做专长之事。\n\n更关键的是训练方法的突破。团队提出了 Flow-GRPO——一种在线强化学习算法。传统方法在智能体与环境离线交互后进行训练，导致学到的策略与真实运行环境脱节。Flow-GRPO 的核心思路是将多轮优化问题分解为一系列可处理的单轮策略更新，将一个可验证的轨迹级最终outcome reward广播回每一轮，帮助规划器在流中完成端到端优化，同时通过组归一化优势函数稳定学习过程。\n\n实验数据显示，7B AgentFlow 在搜索任务上平均提升 14.9%，智能体任务提升 14.0%，数学任务提升 14.5%，科学任务提升 4.1%——不仅超越同级别开源模型，也超越了规模更大的 GPT-4o。随着模型参数规模增大或推理轮次增加，AgentFlow 的优势呈现正 scaling，暗示这一训练范式具有可扩展性。\n\nAgentFlow 的成功揭示了一个重要趋势：智能体的训练范式正从单模型离线训练向多模块在线协同优化演进。过去将推理与工具调用混在一个模型里一条道走到底的思路，可能正在被更具架构意识的模块化方法所取代。对开发者而言，这意味着与其追逐更大的基础模型，不如关注训练框架的革新——同样的7B参数，正确的方法可以带来质的飞跃。","https:\u002F\u002Fopenreview.net\u002Fforum?id=Mf5AleTUVK","ec0a79b7-694c-4caf-8071-91315d69c706",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"d3644372-38a7-4519-a377-dc06e5bbf679","en","AgentFlow: streaming training lets a 7B model top GPT-4o","When the industry is still using \"bigger is better\" to pile up parameters, a paper just presented at ICLR 2026 as an Oral gives a different answer. \"In-the-Flow Agentic System Optimization for Effective Planning and Tool Use,\" published by a multi-institution research team, proposes a new agent architecture called AgentFlow — using only a 7B-parameter base model, it comprehensively surpasses GPT-4o across 10 benchmarks including search, math, and scientific reasoning, with an average lead of 14.9%.\n\nThe core innovation behind this is not a bigger model, but a brand-new modular agent framework and online reinforcement learning training method.\n\nCurrent mainstream Tool-Augmented LLMs commonly use a single monolithic strategy, mixing chain-of-thought and tool calling within a complete context. This monolithic design works when the task chain is short and the variety of tools is limited, but once faced with long-horizon planning or diverse tool calling, the linear growth of context leads to dramatic efficiency decline and weak generalization.\n\nAgentFlow, on the other hand, decomposes the agent into four modules each handling their specialty: Planner (decomposes tasks), Executor (calls tools and returns results), Verifier (checks correctness of intermediate outputs), Generator (memory management and context construction). The four modules coordinate dynamically through evolutionary memory, each doing its specialty work.\n\nMore critical is the training method breakthrough. The team proposes Flow-GRPO — an online reinforcement learning algorithm. Traditional methods train after agents interact with the environment offline, leading to learned policies being out of touch with real running environments. Flow-GRPO's core idea is decomposing the multi-turn optimization problem into a series of manageable single-turn policy updates, broadcasting a verifiable trajectory-level final outcome reward back to each turn, helping the planner complete end-to-end optimization in the flow, while stabilizing the learning process through group-normalized advantage functions.\n\nExperimental data shows 7B AgentFlow achieves an average 14.9% improvement on search tasks, 14.0% on agent tasks, 14.5% on math tasks, 4.1% on science tasks — not only surpassing same-tier open-source models, but also the larger-scale GPT-4o. As model parameter scale or inference turn count increases, AgentFlow's advantage shows positive scaling, hinting at the scalability of this training paradigm.\n\nAgentFlow's success reveals an important trend: agent training paradigms are evolving from single-model offline training to multi-module online collaborative optimization. The previous approach of mixing reasoning and tool calling in a single model, all-the-way-through, may be replaced by more architecturally-aware modular methods. For developers, this means rather than chasing bigger foundation models, focus on training framework innovation — with the same 7B parameters, the right method can bring qualitative leaps.","agentflow-7b-icrl-oral-flow-grpo-14-9pct","2026-05-03T01:10:00Z","2026-05-03T01:08:01.808618Z","2026-08-19T02:08:40.142862Z",true,"agent",120,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"4bbc55d2-cabc-477f-a3ad-4e2c119aff2a","TokTier 抓住 Agent 推理的隐藏瓶颈：缓存命中 94.1%，分词仍吃掉 64% 首 token 时间","toktier-stateful-tokenization-agent-serving","2026-07-31T17:56:30+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"6b52b4a9-d567-46b8-99c1-e9c65ba59b16","SWE-Pruner Pro:ByteDance 让 Agent 自己当剪枝器,省 39% token 还涨分","swe-pruner-pro-bytedance","2026-07-25T12:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"37aa0bc9-d135-444f-842e-0b40388d29e9","Qwen3.7-Max 原生兼容 Anthropic API 协议：Claude Code 现已可直接调用阿里模型","qwen3-7-max-anthropic-api-claude-code","2026-05-27T10:05:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"57b691cc-476c-4427-8618-e29127654b34","AMD ROCm 7 原生支持 Qwen3-Coder-Next：单卡 256k 上下文打破推理硬件垄断","amd-rocm7-qwen3-coder-next-256k-mono","2026-05-25T16:10:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"0d0e5ce8-fa18-4907-b811-2918ff8464e4","FlexSQL：小型LLM如何在Text-to-SQL任务上超越GPT-o3和DeepSeek-R1","flexsql-nus-text-to-sql-spider2-65pct-gpt-oss-120b","2026-05-05T10:15:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"cac485ab-a429-4ebc-88c6-a1f924f978ff","AWS 开源 KeysAndValues:微调时就让模型学会“遗忘”,单张 A100 撑住 128K","aws-keysvalues-sparse-attention-finetuning","2026-08-26T05:20:00+00:00"]