[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-claude-opus-4-7-release-workflow-vision-en":3,"article-related-claude-opus-4-7-release-workflow-vision-en":25,"series-model-release-2ab61916-02e3-47f5-8131-9d69cb770f03":77},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":11,"views":22,"created_at":23,"published_at":24,"topic_cluster_id":11},"2ab61916-02e3-47f5-8131-9d69cb770f03","claude-opus-4-7-release-workflow-vision-en","Claude Opus 4.7 发布：更会干活了","\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002F\" target=\"_blank\" rel=\"noopener\">Anthropic\u003C\u002Fa> 发布 \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-opus-4-7\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7\u003C\u002Fa>，这次升级的重点很明确：复杂任务执行、高清视觉理解、长链路工作流稳定性。官方给出的定位也很直接，它面向的是那些真的要把活做完的场景，而不是只会把答案说得漂亮。\u003C\u002Fp>\u003Cp>这次更新最值得注意的一点，是它把“模型会不会做事”摆到了“模型会不会聊天”前面。对于开发者、分析师、法务、研究人员来说，这种变化比单次跑分更有意义，因为它直接影响交付质量、返工次数和上下文管理成本。\u003C\u002Fp>\u003Cp>如果你平时会让模型改代码、读截图、整理材料、做演示文稿，Opus 4.7 不是那种看一眼参数就能忽略的小版本。它的变化很像一次面向办公场景和代理式工作流的升级，代价也很现实：更高分辨率输入和更长输出，都会更快消耗 Token。\u003C\u002Fp>\u003Ch2>这次升级，重点不在“更会聊”\u003C\u002Fh2>\u003Cp>Anthropic 把 Opus 4.7 的核心能力放在高级软件工程、长时间任务执行和更严格的指令遵循上。简单说，模型不再只是回答问题，而是更像一个能跟着步骤做完任务的执行者。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776859438392-675v.png\" alt=\"Claude Opus 4.7 发布：更会干活了\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>官方 API 说明里，它被描述为当前最强的通用可用模型之一，尤其适合复杂推理和代理式编码。这个方向很清楚：大模型竞争的焦点，已经从“答案像不像人”转向“事情做没做成”。\u003C\u002Fp>\u003Cp>从产品角度看，这意味着用户会更少遇到那种“前半段答得很好，后半段开始跑偏”的情况。对于长文档改写、跨文件整理、代码审查这类任务，稳定性比华丽措辞更重要。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-opus-4-7\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7 官方发布页\u003C\u002Fa>强调复杂任务和长链路执行\u003C\u002Fli>\u003Cli>SWE-bench Multilingual 上，Opus 4.7 得分 80.5%，Opus 4.6 为 77.8%\u003C\u002Fli>\u003Cli>GraphWalks BFS 1M 场景中，Opus 4.7 从 41.2% 提升到 58.6%\u003C\u002Fli>\u003Cli>Vending-Bench 2 中，Opus 4.7 最终余额 10,937 美元，Opus 4.6 为 8,018 美元\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>视觉能力这次补得很猛\u003C\u002Fh2>\u003Cp>这次最容易被普通用户感知到的变化，是它看图更细了。Anthropic 提到，Opus 4.7 支持长边最高 2576 像素的图像输入，约 375 万像素，明显高于此前版本。对密集截图、复杂图表、流程图、产品原型图来说，这种能力提升很实用。\u003C\u002Fp>\u003Cp>过去很多模型在高分辨率界面里容易漏掉小字、按钮和局部结构。Opus 4.7 的变化在于，它更像是把“看得见”变成了“看得清”。这对 Computer Use 场景尤其重要，因为 UI 元素常常只占整张图很小一块面积。\u003C\u002Fp>\u003Cp>在 ScreenSpot-Pro 上，Opus 4.7 的表现也很亮眼。低分辨率且不带工具时，它拿到 69.0%，而 Opus 4.6 是 57.7%。切到高分辨率后，Opus 4.7 不带工具就达到 79.5%，叠加工具调用后升到 87.6%。\u003C\u002Fp>\u003Cblockquote>“The future is already here — it’s just not very evenly distributed.” — William Gibson\u003C\u002Fblockquote>\u003Cp>这句话放在今天的模型升级上很贴切。对一部分人来说，AI 还只是聊天工具；对另一部分人来说，它已经开始接手截图分析、界面定位和文档整理。Opus 4.7 让这个分界线又往前推了一步。\u003C\u002Fp>\u003Ch2>和老对手比，差距开始变得具体\u003C\u002Fh2>\u003Cp>如果只看自家版本迭代，Opus 4.7 只是比 Opus 4.6 更强一点。但把它放到同类模型里，差距就更容易看清。Artificial Analysis 基于 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fgdpval\u002F\" target=\"_blank\" rel=\"noopener\">OpenAI GDPval\u003C\u002Fa> 数据集做的 \u003Ca href=\"https:\u002F\u002Fartificialanalysis.ai\u002Fevals\u002Fgdpval-aa\" target=\"_blank\" rel=\"noopener\">GDPval-AA\u003C\u002Fa> 评估，覆盖 44 种知识工作职业和 9 大行业，任务来自平均 14 年经验的资深从业者。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776859438929-639x.png\" alt=\"Claude Opus 4.7 发布：更会干活了\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>在这项评估里，Opus 4.7 的 Elo 分数是 1753，Opus 4.6 是 1619，\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Findex\u002Fgpt-5-4\u002F\" target=\"_blank\" rel=\"noopener\">GPT-5.4\u003C\u002Fa> 是 1674，\u003Ca href=\"https:\u002F\u002Fwww.google.com\u002Fintl\u002Fen\u002Fai\u002Fgemini\u002F\" target=\"_blank\" rel=\"noopener\">Gemini 3.1 Pro\u003C\u002Fa> 是 1314。这个结果很直白：Opus 4.7 已经把不少“写得像样”但“做不完活”的模型甩在了后面。\u003C\u002Fp>\u003Cp>在企业文档推理场景里，差距更夸张。Databricks 的 \u003Ca href=\"https:\u002F\u002Fwww.databricks.com\u002Fblog\u002Fofficeqa-pro-benchmark\" target=\"_blank\" rel=\"noopener\">OfficeQA Pro\u003C\u002Fa> 测的是近 100 年美国财政部公报，语料有 8.9 万页 PDF 和 2600 万个数字。Opus 4.7 在这里拿到 80.6%，Opus 4.6 是 57.1%，GPT-5.4 是 51.1%，Gemini 3.1 Pro 是 42.9%。\u003C\u002Fp>\u003Cul>\u003Cli>GDPval-AA：Opus 4.7 1753，GPT-5.4 1674，Gemini 3.1 Pro 1314\u003C\u002Fli>\u003Cli>OfficeQA Pro：Opus 4.7 80.6%，Opus 4.6 57.1%，GPT-5.4 51.1%\u003C\u002Fli>\u003Cli>Structural Biology：Opus 4.7 74.0%，Opus 4.6 30.9%\u003C\u002Fli>\u003Cli>SWE-bench Multimodal：Opus 4.7 34.5%，Opus 4.6 27.1%\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>成本和安全，还是绕不开的话题\u003C\u002Fh2>\u003Cp>Opus 4.7 的提升不是白来的。Anthropic 明确提到，更高分辨率图像会消耗更多 Token，新的分词器也会让同样输入产生更多 Token，输出在高 Effort 模式下也会增加。对个人用户来说，这意味着额度可能更快见底；对团队和 API 用户来说，这就是实打实的成本变量。\u003C\u002Fp>\u003Cp>另一个不能忽视的点是安全。Anthropic 在发布前一周公布了 \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fproject-glasswing\" target=\"_blank\" rel=\"noopener\">Project Glasswing\u003C\u002Fa>，讨论前沿模型在网络安全方向的风险和收益。Opus 4.7 是这套思路下第一个公开部署的模型，官方还提到它带有自动检测和拦截高风险网络安全请求的护栏。\u003C\u002Fp>\u003Cp>安全评估里，它和 Opus 4.6 的整体画像接近，在诚实性和抵抗提示词注入方面更强，但部分细项也有小幅波动。Anthropic 的态度很明确：这不是一次把所有风险都抹平的发布，而是一次把能力往前推、同时继续收紧边界的发布。\u003C\u002Fp>\u003Cp>对真正会付费的人来说，这些细节比宣传语更重要。因为你买到的不是“更聪明”，而是“更能干活，但也更吃资源”的模型。\u003C\u002Fp>\u003Ch2>结论：它会先改变谁的工作方式？\u003C\u002Fh2>\u003Cp>最先感受到 Opus 4.7 变化的人，大概率不是普通聊天用户，而是每天要处理代码、表格、截图、文档和长任务流的人。它的价值不在于每次回答都更有文采，而在于更少跑偏、更少返工、更少人工盯着它。\u003C\u002Fp>\u003Cp>我更愿意把这次更新理解成一次工作方式的微调：如果你的流程本来就依赖模型做初稿、做校对、做资料整合，那么 Opus 4.7 会让这条链路更值得信赖；如果你只是偶尔问个问题，体感变化未必有那么强。\u003C\u002Fp>\u003Cp>接下来值得观察的，不是它能不能继续刷高分，而是企业会不会真的把更多中间环节交给它。换句话说，问题已经不是“模型能不能写”，而是“它能不能在你的流程里少出错地写完”。\u003C\u002Fp>","Anthropic发布Claude Opus 4.7，长任务、视觉理解和代码工作流更强，但Token消耗也更高。","zhuanlan.zhihu.com","https:\u002F\u002Fzhuanlan.zhihu.com\u002Fp\u002F2028396026466247335",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776859438392-675v.png","model-release","en","97f9c411-2849-4a70-9d1e-3bba0dde23bf",[17,18,19,20,21],"Claude Opus 4.7","Anthropic","computer use","vision","agentic coding",4,"2026-04-22T12:03:39.271461+00:00","2026-04-22T12:03:39.08+00:00",{"tags":26,"relatedLang":36,"relatedPosts":40},[27,29,31,33,34],{"name":17,"slug":28},"claude-opus-47",{"name":18,"slug":30},"anthropic",{"name":21,"slug":32},"agentic-coding",{"name":20,"slug":20},{"name":19,"slug":35},"computer-use",{"id":15,"slug":37,"title":38,"language":39},"claude-opus-4-7-release-workflow-vision-zh","Claude Opus 4.7 上線：更會做事了","zh",[41,47,53,59,65,71],{"id":42,"slug":43,"title":44,"cover_image":45,"image_url":45,"created_at":46,"category":13},"58aa41ca-2c5f-44c6-ab07-2002473e95b1","gemini-1-5-pro-002-flash-002-2-0-flash-update-en","Gemini 1.5 Pro-002, Flash-002 and 2.0 Flash update Google AI","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780999383257-jccn.png","2026-06-09T10:02:28.362637+00:00",{"id":48,"slug":49,"title":50,"cover_image":51,"image_url":51,"created_at":52,"category":13},"435fc551-a461-444a-bf95-dbf5685cfac0","minimax-m3-open-weight-coding-win-en","MiniMax M3 Proves Open-Weight Can Still Win on Coding","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780968781159-odhi.png","2026-06-09T01:32:31.256895+00:00",{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":13},"12af5a0d-1bbf-4a50-a391-b53f8003f234","gemini-35-flash-pricing-benchmarks-en","Gemini 3.5 Flash Pricing, Context, Benchmarks","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780840981235-e7hm.png","2026-06-07T14:02:30.280485+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":13},"0e767e9d-5d17-4cd0-b6ee-0328f89eb49b","gemma-4-12b-specs-benchmarks-run-locally-en","Gemma 4 12B: Specs, Benchmarks & How to Run It Locally","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780777984661-5ymr.png","2026-06-06T20:32:25.294996+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":13},"9d15f962-739d-44f8-a7f9-11bca64d38e0","best-kimi-models-2026-k2-5-vs-k2-thinking-en","Best Kimi Models in 2026: K2.5 vs K2 Thinking","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780770786284-shy0.png","2026-06-06T18:32:39.779504+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":13},"34547376-5d6b-4453-8d80-8072d8ac36ed","kimi-k2-6-open-source-coding-agent-swarm-en","Kimi K2.6 adds open-source coding and agent swarm","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780761781526-wop4.png","2026-06-06T16:02:22.26883+00:00",[78,83,88,93,98,103,108,113,118,123],{"id":79,"slug":80,"title":81,"created_at":82},"d4cffde7-9b50-4cc7-bb68-8bc9e3b15477","nvidia-rubin-ai-supercomputer-en","NVIDIA Unveils Rubin: A Leap in AI Supercomputing","2026-03-25T16:24:35.155565+00:00",{"id":84,"slug":85,"title":86,"created_at":87},"eab919b9-fbac-4048-89fc-afad6749ccef","google-gemini-ai-innovations-2026-en","Google's AI Leap with Gemini Innovations in 2026","2026-03-25T16:27:18.841838+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"5f5cfc67-3384-4816-a8f6-19e44d90113d","gap-google-gemini-ai-checkout-en","Gap Teams Up with Google Gemini for AI-Driven Checkout","2026-03-25T16:27:46.483272+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"f6d04567-47f6-49ec-804c-52e61ab91225","ai-model-release-wave-march-2026-en","Navigating the AI Model Release Wave of March 2026","2026-03-25T16:28:45.409716+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"895c150c-569e-4fdf-939d-dade785c990e","small-language-models-transform-ai-en","Small Language Models: Llama 3.2 and Phi-3 Transform AI","2026-03-25T16:30:26.688313+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"38eb1d26-d961-4fd3-ae12-9c4089680f5f","midjourney-v8-alpha-features-pricing-en","Midjourney V8 Alpha: A Deep Dive into Its Features and Pricing","2026-03-26T01:25:36.387587+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"bf36bb9e-3444-4fb8-ab19-0df6bc9d8271","rag-2026-indispensable-ai-bridge-en","RAG in 2026: The Indispensable AI Bridge","2026-03-26T01:28:34.472046+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"60881d6d-2310-44ef-b1fb-7f98e9dd2f0e","xiaomi-mimo-trio-agents-robots-voice-en","Xiaomi’s MiMo trio targets agents, robots, and voice","2026-03-28T03:05:08.899895+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"f063d8d1-41d1-4de4-8ebc-6c40511b9369","xiaomi-mimo-v2-pro-1t-moe-agents-en","Xiaomi MiMo-V2-Pro: 1T MoE Model for Agents","2026-03-28T03:06:19.238032+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"a1379e9a-6785-4ff5-9b0a-8cff55f8264f","cursor-composer-2-started-from-kimi-en","Cursor’s Composer 2 started from Kimi","2026-03-28T03:11:59.132398+00:00"]