[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-qwen-3-7-flash":3,"news-related-c329665c-c1fd-4af4-ad28-a1fbf9a8ede7":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},"c329665c-c1fd-4af4-ad28-a1fbf9a8ede7","Qwen3.7-Flash 上线：多模态 Agent 从演示走向 CI 与搜索一线","阿里在 7 月 25 日的 QwenCloud changelog 中悄然放出 qwen3.7-flash 与快照版 qwen3.7-flash-2026-07-15。作为 3.6-Flash 的继承者,这是 Qwen3.7 系列里第一颗定位于视觉-语言快响应赛道的小尺寸模型——重点不是更大的参数量,而是把“看得清、操作得稳”这件事打磨到能进生产环境。\n\n升级点集中在四块:通用目标识别能力提升,真实世界感知与空间智能增强,Search Agent 和 CI Agent 场景下端到端任务执行更稳定,以及为“vibe coding”体验专门优化的多模态编码。Search Agent 的含义很直白——给模型喂搜索结果片段时,它能更准确地抽取实体、判断证据并规划下一步动作;CI Agent 则是把模型塞进持续集成流水线,让它读懂失败日志、看 diff、提补丁,而不是只会聊天。\n\n比较有趣的是,Qwen 这次没把 3.7-Flash 当成纯文本模型的轻量版,而是把它定位成“能在 CI 里跑起来的多模态 Agent”。过去这一类角色通常被闭源旗舰或大尺寸开源模型占据,Flash 体量的多模态小模型介入意味着 Agent 成本和延迟能再下一个台阶——和 Gemini 3.6 Flash、GLM-5.2 的开源节奏形成同一波“小而能打”的趋势。\n\n对开发者来说,值得关注的点不是参数表,而是它和 Qwen3.7-Plus、Qwen3.7-Max 在多模态 Agent 能力上的梯度划分:Max 跑长程、Plus 跑交互、Flash 跑流水线。这条线一旦在 CI 场景里跑通,Agent 就不再是 demo 视频里的“自动修一个 bug”,而是会真的进入每天几千次 PR 的工程循环。","https:\u002F\u002Fdocs.qwencloud.com\u002Fchangelog\u002Fmodels","c36a21ac-2a77-421b-9519-1e150695732a",[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},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"1b927903-83f0-4753-a41c-13c0c6ffdc2e","en","Qwen3.7-Flash: multimodal agents move from demo to CI and search","Alibaba quietly dropped qwen3.7-flash and snapshot version qwen3.7-flash-2026-07-15 in the QwenCloud changelog on July 25. As the successor to 3.6-Flash, it's the first small-sized model in the Qwen3.7 series positioned in the vision-language fast-response lane — the focus isn't bigger parameters, but polishing \"see clearly, act stably\" to a level that can run in production environments. The upgrades concentrate in four areas: general object recognition improved, real-world perception and spatial intelligence enhanced, end-to-end task execution more stable in Search Agent and CI Agent scenarios, and multimodal encoding specifically optimized for \"vibe coding\" experience. The meaning of Search Agent is straightforward — when fed search-result snippets, the model can more accurately extract entities, judge evidence, and plan the next move. CI Agent puts the model into the continuous integration pipeline, letting it read failure logs, look at diffs, propose patches, rather than just chat. What's interesting is that this time Qwen doesn't treat 3.7-Flash as the lightweight version of a pure-text model, but positions it as a \"multimodal Agent that can run in CI\". In the past, this role was usually occupied by closed-source flagships or large-sized open-source models. The intervention of Flash-class multimodal small models means Agent cost and latency can drop another notch — forming the same wave of \"small but capable\" as the Gemini 3.6 Flash and GLM-5.2 open-source rhythm. For developers, what matters isn't the parameter table, but its gradient division of multimodal Agent capability against Qwen3.7-Plus and Qwen3.7-Max: Max runs long-horizon, Plus runs interaction, Flash runs pipeline. Once this line proves itself in CI scenarios, the Agent will no longer be a \"fix one bug\" demo in a video, but will really enter the daily multi-thousand-PR engineering loop.","qwen-3-7-flash","2026-07-28T02:02:00Z","2026-07-27T18:04:46.781118Z","2026-08-19T02:08:40.142862Z",true,"agent",162,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"cb64371f-62b6-473d-8150-b576001d3f56","Qwen3.8-27B 开源权重上线:单卡跑得动的 Qwen3.8,还塞了个视觉编码器","qwen3-8-27b-open-weights-release","2026-08-14T19:30:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"40095b51-97b0-4fd4-9b1d-f636c970572e","阿里 Qwen 团队发布 Qwen3.8-Max:2.4 万亿参数 MoE 模型首度开放权重","qwen3-8-max-2-4t-moe-open-weights","2026-08-07T02:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"7258978b-dfcd-4cb4-91c4-3b8569cd5deb","Qwen-Audio-3.0-TTS双版本发布:Plus登顶Artificial Analysis,Flash压到300ms首包延时","qwen-audio-3-tts","2026-07-20T10:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"73545307-3c18-48ab-aac9-5a4c7fc96b53","Meta Muse Spark 1.1：1M 上下文，Computer Use 自主决策","meta-muse-spark-1-1","2026-07-09T16:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"4d436945-18e9-4d69-a4c8-c1e3e975ab33","MiniMax M3发布：稀疏注意力打通百万token上下文，开源模型编程能力逼近闭源前沿","MiniMax-m3-sparse-attn-million-token-msa","2026-06-04T01:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"4c7f5330-3aff-458a-9ef5-f04cc5585703","微信视觉团队开源 WeMM 嵌入模型:2B 反超 8B 前基线,9B 达 MMEB-v2 80.6","wemm-embedding-wechat-multimodal","2026-08-26T21:07:30+00:00"]