[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-terminalhorizon-long-trajectory-27b-agent":3,"topics-all":38,"news-related-fead7184-42cc-4604-9aba-48dcb20ef7ab":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"fead7184-42cc-4604-9aba-48dcb20ef7ab","阿里2852条长轨迹:喂出12基准全涨的27B智能体","高德与中科大团队发布 TerminalHorizon 数据合成引擎,用 2852 条长时程轨迹微调 Qwen3.5-27B,12 项基准全部上涨,Terminal-Bench 2.1 从 46.82 到 60.67,真实自动化任务得分接近翻倍,点云任务里模型持续研究 2.21 小时不弃场。","终端智能体正在从「帮你在本机敲几条命令」走向「独立完成专业工作」,但训练数据成了新的瓶颈:现有合成环境产出的轨迹大多是一次性的短操作——复制文件、装依赖、跑个测试——智能体没机会练习真正值钱的东西:探索取证、根据反馈修改策略、把中间结果攒成最终交付物。阿里高德(AMAP)团队与中国科学技术大学的新工作 TerminalHorizon 就是冲着这个缺口去的。\n\n## 合成数据的盲点:环境把活儿提前干完了\n\n论文指出的问题很尖锐:环境合成常常在构建阶段就把智能体该做的关键工作预先完成或简化掉,探索、实现、验证被「剧透」,轨迹自然教不出完整的问题解决过程。TerminalHorizon 的解法分两步:先用独立执行探测环境构建时漏掉了哪些关键工作,再拿这些证据反向指导环境重建;随后对同一目标渐进扩展任务,不断引入依赖先前成果的新需求,逼着轨迹长出跨阶段的连续性。\n\n产出规模:1535 个可执行环境,覆盖 25 个专业领域、463 个子域(从软件工程、科学计算到农业、公共事务),归纳出探索、诊断、构建、验证、协调五大族共 27 种行为模式。数据的「长」是实打实的:平均交互 52.82 轮(P50 为 40 轮),对照次高的 CLI-Gym 平均只有 15.60 轮。\n\n## 12 项基准全部上涨\n\n团队用 2852 条轨迹对 Qwen3.5-27B 做监督微调,得到 TerminalHorizon-27B,在 12 项基准上全部提升:Terminal-Bench 1.0 从 46.3 到 55.0,2.0 从 41.6 到 55.8,2.1 从 46.82 到 60.67;涨幅最大的是真实生产环境任务 AutomationBench,partial 指标从 26.4 涨到 49.7(+23.3 个百分点),通过率从 5.5 到 17.2,翻了三倍多。通用推理也在涨(AIME 2025 从 86.7 到 93.3),只是幅度相对温和。\n\n两个附加实验值得注意。其一是数据规模 scaling:1000 条轨迹时 TB 2.1 已到 57.30,2852 条到 60.67,仍有上探空间;其二是配方可迁移:同一套数据微调 35B-A3B 与 122B-A10B 两个更大基座,TB 2.1 分别从 39.70 涨到 52.81、49.06 涨到 64.04。\n\n## 12 小时自主研究:学会不放弃\n\n最有意思的是 RSI(递归自我改进)案例:每例给 12 小时研究预算,看模型能否在初始方案后继续实验、按反馈改进、把有效改进留在交付物里。以多视角点云融合任务为例,基座 Qwen3.5-27B 在 12.1 分钟后主动停止,最终 F-score 53.62;TerminalHorizon-27B 持续迭代到 2.21 小时才锁定最终代码,F-score 达到 88.12,隐藏奖励从 21.93 涨到 48.96。「不轻易放弃」本身,成了长轨迹数据教出来的能力。\n\n## 冷水与看点\n\n这条赛道今年已经很拥挤:Qwen 的 Terminal-Universe 把旧轨迹反向重建为 3.73 万个环境,FACET 先建环境再写任务,都聚焦「环境从哪来」;TerminalHorizon 押的是另一个轴——轨迹里工作的完整性。12 项全涨、真实生产任务翻倍级的提升,说明「长」确实是有效信号。冷水也要泼:方法只验证了 SFT 路线,通用推理涨幅有限,数字全部来自团队自报,尚无第三方复现。\n\n对一个由地图业务团队牵头的工作来说,产业意图很明显:高德场景里沉淀着大量真实的专业级长时程任务,这是别家没有的数据腹地。当同行还在卷环境合成技巧时,「谁手里有真实专业工作流」可能才是终端智能体下一阶段的分水岭。论文地址:https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13166","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13166","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"5a98b8e4-317f-426d-8e05-a291fb067641","en","2,852 Long Trajectories Lift a 27B Agent on All 12 Benchmarks","AMAP's TerminalHorizon fine-tunes Qwen3.5-27B on 2,852 long trajectories, gaining on all 12 benchmarks with Terminal-Bench 2.1 up from 46.82 to 60.67.","Terminal agents are moving from \"run a few commands on your machine\" toward completing professional work independently. The new bottleneck is training data: most synthetic environments produce short, one-shot trajectories — copy a file, install a dependency, run a test — leaving no room for the skills that actually matter: gathering evidence through exploration, revising strategy from feedback, and turning intermediate results into final deliverables. TerminalHorizon, from Alibaba's AMAP team and the University of Science and Technology of China, targets exactly this gap.\n\n## The Blind Spot in Synthetic Data: Environments Do the Work in Advance\n\nThe paper is blunt about the problem: environment synthesis often completes or simplifies the key work during construction, \"spoiling\" the exploration, implementation, and verification the agent was supposed to practice. TerminalHorizon's answer has two stages. First, independent execution probes what key work was omitted when the environment was built, and the resulting evidence guides reconstruction. Second, tasks under the same goal are progressively expanded, introducing new requirements that build on prior results — forcing trajectories to grow cross-stage continuity.\n\nThe output: 1,535 executable environments covering 25 professional domains and 463 subdomains, from software engineering and scientific computing to agriculture and public services, organized into 27 behavioral patterns across five families (explore, diagnose, build, verify, coordinate). The \"long\" in long-horizon is literal: mean interaction length is 52.82 turns (P50 of 40), versus 15.60 for the next-highest method, CLI-Gym.\n\n## All 12 Benchmarks Improve\n\nThe team supervised-fine-tuned Qwen3.5-27B on 2,852 trajectories to produce TerminalHorizon-27B, which improves on all 12 benchmarks: Terminal-Bench 1.0 from 46.3 to 55.0, 2.0 from 41.6 to 55.8, and 2.1 from 46.82 to 60.67. The largest jump comes on AutomationBench, which measures agentic tasks in real production settings: the partial metric climbs from 26.4 to 49.7 (+23.3 points), and the pass rate goes from 5.5 to 17.2 — more than tripling. General reasoning also rises (AIME 2025 from 86.7 to 93.3), though more modestly.\n\nTwo additional experiments stand out. On data scaling, Terminal-Bench 2.1 already reaches 57.30 with 1,000 trajectories and 60.67 with 2,852 — headroom remains. And the recipe transfers: fine-tuning two larger bases (35B-A3B and 122B-A10B) on the same data lifts TB 2.1 from 39.70 to 52.81 and 49.06 to 64.04 respectively.\n\n## 12-Hour Autonomous Research: Learning Not to Give Up\n\nThe most striking part is the RSI (recursive self-improvement) case studies: each case grants a 12-hour research budget to test whether a model can keep experimenting beyond an initial solution, improve its methods from feedback, and retain effective improvements in its deliverable. On a multi-view point-cloud fusion task, the base Qwen3.5-27B stops voluntarily after 12.1 minutes with a final F-score of 53.62; TerminalHorizon-27B keeps iterating until it locks its final code at 2.21 hours, reaching an F-score of 88.12, with the hidden reward rising from 21.93 to 48.96. \"Not giving up easily\" turns out to be a skill that long trajectories teach.\n\n## Caveats and What to Watch\n\nThe lane is crowded this year: Qwen's Terminal-Universe reverse-rebuilds old trajectories into 37,300 environments, and FACET builds environments before writing tasks — both focused on where environments come from. TerminalHorizon bets on a different axis: the completeness of the work inside each trajectory. All-12 gains and doubling-level improvements on production tasks suggest \"long\" is a real signal. The cold water: only the SFT recipe is validated, general-reasoning gains are limited, and every number is self-reported with no third-party replication yet.\n\nFor a work led by a mapping-business team, the industrial intent is clear: AMAP's services hold deep reserves of real professional long-horizon tasks — a data moat few others have. While the field competes on environment-synthesis tricks, \"who holds real professional workflows\" may be the dividing line for the next stage of terminal agents. Paper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.13166","terminalhorizon-long-trajectory-27b-agent","2026-10-13T21:08:01Z","2026-10-13T21:09:48.309356Z","2026-10-13T21:09:48.309371Z",true,"agent",17,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"d552a77b-3b52-4d59-8962-e897aa06b37c","腾讯RSR:把借来的轨迹改写成自己的能力","recursive-self-rewrite-terminal-trajectories","2026-10-06T21:06:49+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"c77381d9-29ba-45ff-89df-855d11f90de2","Terminal-Universe:Qwen 把旧轨迹反向重建为 3.73 万个环境,27B 微调双基准 +11.9\u002F+13.8 分","terminal-universe-trajectory-environments","2026-09-04T17:10:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"859242dd-0035-41e6-8db9-4583989048b7","AgentSWE基准:让编码智能体造智能体,无人及格","agentswe-coding-agent-benchmark","2026-10-13T19:08:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"feb04a1c-2c75-42d2-9b62-9c4f0585d924","AutoCompact:让编码代理自己学会何时压缩上下文","autocompact-learned-context-compaction","2026-10-04T17:10:39+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"206ea36a-1eea-463f-a241-1e3b32f5ec2d","USTC GraphForge:证据图把任务和 rubric 钉在一起,Qwen3.6-27B 涨三基准","graphforge-ustc-qwen36-27b-evidence-graph","2026-10-04T03:05:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"6f4d1046-ee70-4a06-9261-2cc187c66285","12 万美元 token 把 Copilot 运行时搬进 Rust:AI 智能体包揽 43 万行移植","copilot-runtime-rust-agentic-port","2026-09-20T19:11:22+00:00"]