[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-anthropic-claude-code-effort-level":3,"topics-all":36,"news-related-7ca1f9d4-e3e3-48e3-bfb2-ee5a9e6d5176":55},{"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},"7ca1f9d4-e3e3-48e3-bfb2-ee5a9e6d5176","Anthropic 拆开 Claude Code：别再只换模型，把\"努力度\"也调对","7 月 7 日，Anthropic Claude Code 团队成员 Lydia Hallie 写了一篇长文，首次把 Model 与 Effort 两个滑块的底层机制彻底摊开。\n\n**Model 决定\"会不会\"，Effort 决定\"肯不肯干满\"**\n\nModel 切换本质是换一套冻结的权重：训练结束后几十亿参数只读，prompt 和 CLAUDE.md 都改不动它。Effort 则控制 Claude 一次任务上投入多少工作量——读几个文件、跑不跑测试、要不要把多步任务推到底再回来。官方图示：同一条 prompt，高 Effort 路径生成的 token 约为低 Effort 的 7 倍，多出来全花在验证上。\n\n文章用三个比喻：Sonnet 是给你一整个下午的全能选手，Fable 是看一眼就能揪出别人没发现毛病的专科医生，Opus 是只有 5 分钟但经验丰富的专家。结论反常识：中等模型配高 Effort，干翻旗舰模型开低 Effort 并不奇怪。\n\n**三月\"变笨\"风波的真凶**\n\n3 月 4 日，Anthropic 为压延迟把 Effort 默认从 high 改成 medium，更新日志写过却没人在意。AMD AI 负责人 Stella Laurenzo 翻 6852 个会话日志，实测思考量比 2 月跌 67%，撂下\"Claude 已无法被信任干复杂工程活\"。一个月后才调回去。多数人还停在\"换更大的模型\"的老思路里，对手边这个 Effort 开关浑然不觉。\n\n**所以呢**\n\n文章表面教调参，背后指向范式拐点：AI 编程的竞争正从\"谁的模型更强\"转向\"谁更会调度\"。简单改动用 Sonnet 挂低档，大型重构用 Opus 加高档，长跑 Agent 任务用 Fable 配足 Effort。Claude Code 新加的 ultracode 档，等于把这套调度写进产品——选中它，Claude 会自己掂量要不要拉起子 Agent 队伍并行干。\"挑模型\"这个最核心的动作，正在被拆成\"花多少钱 × 派多少活\"两个独立维度；下一个分水岭，会出现在哪个 Agent 框架先把这种调度自动化上。","https:\u002F\u002Fclaude.com\u002Fblog\u002Fclaude-model-and-effort-level-in-claude-code","1fa87d30-d9f3-4752-b3be-0373933b3aaf",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"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},"015b63ce-7976-4221-8496-b360f22831e9","en","Anthropic on Claude Code: tune effort, not just the model","On July 7, Anthropic Claude Code team member Lydia Hallie wrote a long post, for the first time fully laying out the underlying mechanism of the Model and Effort two sliders. **Model decides \"can it\", Effort decides \"is it willing to do its full work\"**. Switching Model is essentially swapping a set of frozen weights: after training ends, billions of parameters become read-only, and neither the prompt nor CLAUDE.md can change it. Effort controls how much work Claude puts into a single task — how many files to read, whether to run tests, whether to push a multi-step task all the way through and come back. The official illustration: on the same prompt, the high-Effort path generates about 7× the tokens of the low-Effort one, all the extras spent on verification. The post uses three analogies: Sonnet is an all-rounder who gives you an entire afternoon, Fable is a specialist who can spot what others miss at a glance, Opus is a veteran expert with only 5 minutes. The conclusion is counter-intuitive: a mid-tier model with high Effort can often beat a flagship model on low Effort. **The real culprit behind March's \"got dumber\" storm**. On March 4, Anthropic changed the Effort default from high to medium to control latency, the changelog wrote it but no one paid attention. AMD AI head Stella Laurenzo flipped through 6,852 session logs, measured a 67% drop in thinking amount compared to February, and dropped \"Claude can no longer be trusted to do complex engineering work\". It wasn't reverted until a month later. Most people are still stuck on the old thinking of \"switch to a bigger model\", completely unaware of the Effort switch sitting right next to them. **So what**. On the surface, the post is about parameter tuning; behind it is a paradigm shift: AI programming competition is shifting from \"whose model is stronger\" to \"who is better at scheduling\". Simple changes use Sonnet at a low tier, large refactors use Opus at a high tier, long-running Agent tasks use Fable with full Effort. The new ultracode tier in Claude Code is essentially writing this scheduling into the product — select it, and Claude itself will judge whether to spin up sub-Agent teams to work in parallel. The most central action, \"pick a model\", is being split into two independent dimensions: \"how much money × how much work\"; the next watershed will appear at whichever Agent framework automates this scheduling first.","anthropic-claude-code-effort-level","2026-07-12T07:00:00Z","2026-07-12T08:05:21.367909Z","2026-08-19T02:08:40.142862Z",true,"agent",183,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"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":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"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":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"bba970e8-e80c-4a7f-9c8c-f4acfe00fbea","Anthropic 开发者大会直击：Code with Claude 揭示编程未来","anthropic-code-with-claude-dreaming","2026-05-21T22:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"390c2437-4e4f-45ec-8270-67c5bfa4fa47","ChatGPT、Claude、Grok、Gemini 罕见同时下线,周四早晨全球 AI 集体失声","chatgpt-claude-grok-gemini-thursday-outage","2026-09-05T06:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"f3d17d45-e1a8-4a1b-9449-6813aff06e49","Anthropic 让 Claude 自己修对齐:10 类失败全部见效,还超过人类研究员","claude-automated-alignment-researchers","2026-08-29T13:05:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"39724847-fdc9-4199-ac46-311e7b49d385","Ramp 数据复盘 Fable 5:旗舰上市两月仅占企业 Anthropic 支出 11%,70 倍价差压住前沿模型溢价","ramp-data-fable-5-adoption-plateaus","2026-08-26T08:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"e1724d68-bf0d-4b3f-8047-147796d5d52e","Ramp 8 月指数:Fable 5 企业份额停滞 11%,OpenAI 旗舰跑赢两倍","anthropic-fable-5-plateau-11-percent","2026-08-25T06:00:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":81},"1051d676-8ed9-4448-b0d5-8db4b844f41f","Claude Fable 5 上线两个月,为什么企业只把 11% 的账单花给最强模型","claude-fable-5-11-percent-anthropic-spend"]