[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-mai-gym-gui-agent-online-rl-9b":3,"topics-all":38,"news-related-aa718f3c-14cf-4984-8762-794eaaee8c16":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},"aa718f3c-14cf-4984-8762-794eaaee8c16","MAI-Gym:9B 智能体在 2 万个并行环境里做在线 RL","阿里通义 MAI 开源 MAI-Gym:把 GUI 智能体的评测、数据采集和在线 RL 训练并进同一条流水线,支持十个基准、扩展到两万并行环境实例。平台自动合成 mock 应用与跨 App 世界,发布 49 个模拟应用和 2000 条执行验证任务。基于它训练的 9B 模型在 MobileWorld 拿到 78.4%。","过去两年，GUI 智能体（让模型直接看屏幕、点按钮、敲键盘的 Agent）一直被两件事卡住：交互环境太贵，以及像样的训练任务太少。前沿模型在数学和代码上已经能靠\"可验证奖励\"吃到大规模在线强化学习的红利，但这套打法搬到 GUI 上就失灵了——你没法让数学答案自动判对错，也很难在几十台真机上并行跑出几十万条轨迹。\n\n阿里通义 MAI 团队这次放出来的 MAI-Gym，试图把评测、数据采集和训练三件事焊进同一条流水线里，一次解决这两道坎。\n\n## 关键设计：评测和训练共用一条 rollout 流水线\n\n这是整套平台最核心的取舍。如果评测用的执行环境和训练用的执行环境是两套代码，那你今天测出来的分数，和明天训出来的模型，中间就隔着一层永远对不齐的偏差。MAI-Gym 把两者合并成同一条 rollout pipeline，让智能体在评测和训练里的行为保持一致。\n\n在这个基础上，平台支持十个既有基准，并能把并行环境实例扩展到 20,000 个。对做 Agent 训练的人来说，这个数字比听上去重要得多——在线 RL 的强度直接取决于你能同时开多少个交互环境。\n\n## 不只是基础设施：自动造 App 和任务\n\n光有并行环境还是不够，因为真实环境的任务标注极其昂贵。所以平台还做了一件事——自动合成 GUI 应用和困难任务，包括可重置的 mock 应用，以及能让任务跨多个 App 保持连贯性的\"跨 App 世界\"，并且配上了基于执行的任务验证器（verifier），任务是否完成由实际跑一遍来判定，而不是靠人写标签。\n\n据论文介绍，任务生成过程会参考模型的弱点来定向出题。发布内容包含 49 个模拟应用、2,000 条经执行验证的任务，以及可复用的合成技能（synthesis skills）。\n\n## MAI-Gym-9B：一个模型管手机和桌面\n\n用这套平台做联合在线 RL 训练，团队得到一个统一的 9B 模型 MAI-Gym-9B，同时处理移动端和桌面端操作。论文报告它在 MobileWorld 上取得 78.4%、在 OSWorld-Verified 上取得 76.2%，并称这是 10B 参数以下开放 GUI 智能体中的领先水平——这个说法来自论文作者自己的评测，其他团队还没有独立复现过。\n\n更有意思的一个数字来自迁移实验：在一个 mock 的火车票订购 App 上做 RL 训练，策略迁移到真实手机上的对应 App 后，任务成功率从 28.8% 提升到 48.1%。这说明 mock 环境造出来的训练信号，确实能穿过仿真与现实之间的缝隙。\n\n## 我的看法：GUI Agent 的瓶颈正在从模型转向环境工程\n\n过去两年的行业叙事几乎都压在模型侧——参数、上下文长度、多模态能力。但 GUI Agent 真正的落地卡点，正在悄悄移到另一侧：环境怎么搭、任务怎么验、并行度能堆到多少。\n\nMAI-Gym 把这些工程问题打包成一个开源平台，连模型权重一起放出来，意味着任何人搭一套自己的 Agent 流水线时，不必从零开始抠这些细节。当然，模拟环境到真实世界的迁移仍然是最大的不确定性——28.8% 到 48.1% 说明迁移确实有效，但也说明迁移之前的效果相当有限，一半以上的任务在真机上直接失败。\n\n所以这条新闻的看点不只是\"又一个 9B 模型跑分\"，而是开源社区第一次拿到了一个把评测、合成、训练、验证串起来的完整闭环。论文、代码和权重都已放出（详见 [Hugging Face 论文页](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2610.20939)），接下来值不值得关注的，是有多少人能在这个底座上跑出自己的智能体。\n\n当\"可验证奖励\"从数学题蔓延到每一次点击和每一次滑动，Agent 训练的规模化才算真正开始。","https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2610.20939","b0060acd-5f64-427d-9673-5caea2940874",[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},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"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},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"49665270-d0d3-4367-846c-caeeaa7bd5c2","en","MAI-Gym trains a 9B GUI agent across 20,000 parallel environments","Alibaba Tongyi MAI open-sources MAI-Gym, merging GUI agent evaluation, data collection and online RL training into one rollout pipeline that supports ten benchmarks and scales to 20,000 parallel environment instances. The platform auto-synthesizes mocked apps and cross-app worlds, releasing 49 simulated apps and 2,000 execution-verified tasks. The 9B model trained on it scores 78.4% on MobileWorld.","For the past two years, GUI agents — the kind that look at a screen and click buttons — have been held back by two things: interactive environments are expensive, and realistic training tasks are scarce. Frontier models have already harvested the gains of online reinforcement learning with verifiable rewards in math and code, but that playbook breaks down on GUI tasks. You can't auto-grade a math proof the way you can grade an arithmetic answer, and it's hard to run hundreds of thousands of trajectories across dozens of real phones at once.\n\nMAI-Gym, released by Alibaba's Tongyi MAI team, tries to fuse evaluation, data collection, and training into a single pipeline so both bottlenecks get addressed at once.\n\n## The key design: one rollout pipeline shared by eval and training\n\nThis is the platform's most consequential trade-off. If the environment you evaluate in and the environment you train in are two separate codebases, then a score you measure today and a model you train tomorrow are always separated by a layer of drift that can never be reconciled. MAI-Gym collapses both into one rollout pipeline, so the agent behaves consistently across evaluation and training.\n\nOn top of that, the platform supports ten established benchmarks and scales to 20,000 parallel environment instances. For anyone doing agent training, that number matters more than it first appears — the strength of online RL is a direct function of how many interactive environments you can run at once.\n\n## Not just infrastructure: synthesizing apps and tasks\n\nParallel environments alone aren't enough, because annotating realistic tasks is brutally expensive. So the platform does something else — it automatically synthesizes GUI applications and hard tasks, including resettable mocked apps and coherent cross-app worlds where a task stays consistent across multiple applications, backed by execution-validated task verifiers. Whether a task succeeded is decided by actually running it, not by a human-written label.\n\nThe paper notes that task generation is guided by the model's weaknesses. The release includes 49 simulated apps, 2,000 execution-verified tasks, and reusable synthesis skills.\n\n## MAI-Gym-9B: one model for phone and desktop\n\nTrained on this platform with joint online RL, the team produced a unified 9B model, MAI-Gym-9B, that handles both mobile and desktop operation. The paper reports 78.4% on MobileWorld and 76.2% on OSWorld-Verified, and claims state-of-the-art results among open GUI agents under 10B parameters. That claim comes from the authors' own evaluation; no independent team has reproduced it yet.\n\nA more interesting number comes from the transfer experiment: RL trained inside a mocked train-ticketing app improved real-phone task success on its live counterpart from 28.8% to 48.1%. That suggests training signal synthesized in a mock environment does cross the gap between simulation and reality.\n\n## My take: the GUI agent bottleneck is shifting from models to environment engineering\n\nThe past two years of industry narrative has almost entirely focused on the model side — parameters, context length, multimodal capability. But the real blocker for GUI agents is quietly moving to the other side of the equation: how you build environments, how you verify tasks, how high you can scale parallelism.\n\nMAI-Gym packages those engineering problems into one open platform and releases the weights alongside it, so nobody has to rebuild those details from scratch to stand up their own agent pipeline. Simulation-to-real transfer remains the biggest open question — 28.8% to 48.1% shows transfer genuinely works, but it also shows how limited things were beforehand: more than half the tasks failed outright on a real phone.\n\nSo the interesting part of this release isn't \"another 9B model posting numbers.\" It's that the open-source community now has a complete loop joining evaluation, synthesis, training, and verification. Paper, code, and weights are all out (details on the [Hugging Face paper page](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2610.20939)). What's worth watching is how many people can build their own agents on this foundation.\n\nWhen verifiable rewards spread from math problems to every click and every swipe, agent training has only just begun to scale.","mai-gym-gui-agent-online-rl-9b","2026-10-21T09:30:00Z","2026-10-21T09:06:35.909649Z","2026-10-21T09:06:35.909659Z",true,"agent",81,[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},"bb6e0f7b-3899-4e45-a267-bfa512d470b0","FDE-Gym:95个真实交付任务,最强Agent只过14个","fde-gym-agent-delivery-benchmark","2026-10-18T23:08:06+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"4977d1f4-c8f7-480c-aaa1-ec01d69f44d8","8B 拿 SFT+GRPO 打 685B MoE:KaliBench 把 LLM 网络安全工具调用拆到命令行级","kalibench-cybersecurity-cli-runtime-rewards","2026-10-03T05:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"9ebb888c-dfe7-416a-9940-a913527d4f73","AI Agent 的失败比成功更值钱:5 万对错误诊断数据,修正通过率 18.4%→51.1%","agent-error-dataset","2026-10-01T15:11:08+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"e98cf9a2-8348-4142-9a56-c11166774798","SpeakerMem-R1:多方对话记忆,分清谁说了什么","speakermem-r1-multi-party-memory","2026-09-24T19:05:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"089195fb-7fe5-4ba9-a4bc-8e356fe5e923","BAAI把1000个GitHub仓库蒸馏成5000个技能,科研agent奖牌率31%冲到73%","baai-disco-repo-to-skill-library","2026-09-03T17:07:35+00:00"]