Anthropic指南：如何根据任务选择Claude模型与调整Effort
Which Claude model should you build with? Fable, Opus, Sonnet, or Haiku
开发者必读的实操指南，清晰梳理了Claude四款模型的适用场景与性价比权衡，直接指导工程落地选型。
Which Claude model to pick, what they cost, and how to use them together.
如何选择 Claude 模型、它们的定价以及如何使用它们协同工作。
- Author: Addy Osmani (Member of Technical Staff) - Published: 2026-11-09 - URL: <https://claude.dev/blog/which-claude-model-should-you-build-with/>
- 作者：Addy Osmani（技术团队成员） - 发布日期：2026-11-09 - URL：<https://claude.dev/blog/which-claude-model-should-you-build-with/>
The current Claude lineup has four generally available models: Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.2. Claude Mythos 5.2 is our most capable model for cybersecurity and life sciences research, available through our verification programs.
当前的 Claude 产品线有四款通用可用模型：Claude Haiku 5.5、Claude Sonnet 5.5、Claude Opus 5.5 和 Claude Fable 5.2。Claude Mythos 5.2 是我们用于网络安全和生命科学研究的最高能力模型，可通过我们的验证计划获取。
If you build on the Claude Platform or work in Claude Code, you may be wondering which model to use for your common tasks. This article will help answer that question and how to think about the trade-offs.
如果您在 Claude Platform 上构建应用或在 Claude Code 中工作，您可能正在考虑针对常见任务使用哪个模型。本文将帮助您回答这个问题，并探讨如何权衡利弊。
Video: One photo of an alley, rendered in code by Claude Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.2. Each image is a Python program the model wrote, calling a drawing engine it also wrote, shown from first layer to finished. Demo by Lance Martin. — One photo of a narrow alley at night beside four renderings of it in code by Haiku 5.5, Sonnet 5.5, Opus 5.5 and Fable 5.2, each built up from first layer to finished.
视频：一张小巷照片，由 Claude Haiku 5.5、Sonnet 5.5、Opus 5.5 和 Fable 5.2 以代码形式渲染。每张图片都是模型编写的 Python 程序，调用了它自己编写的绘图引擎，展示了从第一层到成品的过程。演示由 Lance Martin 提供。——一张夜间狭窄小巷的照片，旁边配有 Haiku 5.5、Sonnet 5.5、Opus 5.5 和 Fable 5.2 以代码形式渲染的四张图像，每张都是从第一层逐步构建至成品。
TL;DR
TL;DR（太长不看版）
The quick answer is that you should start most work on Opus 5.5 at its default effort level, `medium`.
快速答案是，您应在默认努力级别 `medium` 下，将大部分工作放在 Opus 5.5 上开始。
There's a reason it's the default model in Claude Code. It costs half as much per input and output token as Fable 5.2, and it's the model we recommend starting with for most coding and knowledge work.
它在 Claude Code 中成为默认模型是有原因的。它的每个输入和输出 token 成本仅为 Fable 5.2 的一半，并且它是我们要为大多数编码和知识工作推荐的首选模型。
From there:
在此基础上：
- Move down to Sonnet 5.5 for well-scoped everyday work, like a bug fix or a feature built against a spec. Just because you have a clear spec doesn't mean you should use Sonnet, though. A migration or a performance optimization can have clear specs and still be difficult enough to warrant Opus 5.5, or Fable 5.2 for a major one. - Move further down to Haiku 5.5 for the simplest high-volume work, such as classification, routing, and extraction. Haiku 5.5 is also good to use with subagents that work on a larger model's side tasks. - Move up to Claude Fable 5.2 for the hardest technical projects in coding, finance and the sciences. Also move up when Opus 5.5 at `high` still falls short, or for long unattended runs.
- - 对于范围明确且日常的工作（如修复 bug 或根据规范构建功能），可降级到 Sonnet 5.5。不过，即使有明确的规范，也不意味着一定要使用 Sonnet。迁移或性能优化可能有清晰的规范，但仍然足够困难，值得使用 Opus 5.5，或者对于重大变更则使用 Fable 5.2。 - 进一步降级到 Haiku 5.5 用于最简单的海量工作，例如分类、路由和提取。Haiku 5.5 也适合与处理大型模型辅助任务的子代理配合使用。 - 对于编码、金融和科学领域中最复杂的技术项目，可升级到 Claude Fable 5.2。当 Opus 5.5 在 `high` 级别仍显不足，或需要长时间无人值守运行时，也应升级。
Before you move models, try adjusting your effort level. It's a finer dial for both cost and capability. Past `high`, one more effort level can cost as much as a bigger model, so compare both on your own work. If you're past `high`, test the next model up at a lower effort instead. Conversely, one level down can get you the result you need at a lower rate.
在更换模型之前，请先尝试调整你的努力级别（effort level）。这是一个更精细的旋钮，可同时调节成本与能力。超过 `high` 后，再提升一个努力级别的成本可能等同于使用更大的模型，因此请根据你的实际工作情况进行比较。如果你已处于 `high` 级别以上，不妨将下一个更高阶的模型放在较低的努力级别下测试。反之，降低一个努力级别也可能以更低的速率满足你的需求。
Higher effort helps when the model needs to think longer, call more tools, or check its own work more. Just keep in mind that it also adds latency. Additionally, moving to a higher effort level won't give a model knowledge or skill it doesn't have. For that, you still need to move up a model.
当模型需要更长时间的思考、调用更多工具或更仔细地检查自身工作时，提高努力级别会有所帮助。但请注意，这也会增加延迟。此外，提升到更高的努力级别并不能赋予模型其本身不具备的知识或技能。要实现这一点，你仍需升级到更强大的模型。
Pick a model by task
按任务选择模型
| Task | Starting model | Starting effort | Why | ||
|---|---|---|---|---|---|
| Everyday engineering and debugging. Multi-PR feature work | Opus 5.5 | `medium` (default), then `high` | Balance of cost and capability for everyday tasks. | ||
| Classification, routing, tagging, and high volume extraction | Haiku 5.5 | `medium` (default) | Lowest cost and latency in the model family. | ||
| Subagents that search files, read logs, or summarize | Haiku 5.5 or Sonnet 5.5 | The model's default | Wrong answers are cheap to spot, and the main model has the judgment calls anyway. | ||
| Chat and interactive sessions | Sonnet 5.5 (or Haiku 5.5 for simple chat at volume) | `low` or `medium` | Sonnet 5.5 is fast, especially on `low` or `medium` effort. | ||
| Well-scoped coding, like bug fixes, feature iteration, and spec driven work | Sonnet 5.5 | `medium`, then `high` for harder tasks | It handles scoped tasks well at half the Opus 5.5 token price. | ||
| Everyday documents, slides, and spreadsheets | Sonnet 5.5 | `high` (default; `medium` in Claude Code) | It iterates fast and has a strong eye for design. For a complex financial model, or a document built from a large pile of source material, start on Opus 5.5. | ||
| Agent tasks you run repeatedly such as investigation, review, or drafting | Sonnet 5.5 | `medium` | Well-defined, repeated tasks rarely need a bigger model. | ||
| Computer use and vision-heavy workflows | Opus 5.5 | `medium` | These runs are long agentic loops over many screenshots and images, the kind of long-running agentic work Opus 5.5 is built for. | ||
| Major refactors and migrations, architecture planning, and leading many subagents | Fable 5.2 | `medium`, then `high` (the API default is `high`) | It's Anthropic's most capable generally available model for technical fields. | ||
| Deep research, and scientific or quantitative analysis | Fable 5.2 | `medium`, then `high` (the API default is `high`) | It's built for the hardest technical projects in coding, finance and the sciences. | ||
| You aren't sure yet | Opus 5.5 | `medium` (default) | Claude Code uses it by default on Pro, Max, Ultra, Team, Enterprise and the Anthropic API, and the platform docs recommend it as the starting point. |
| 任务 | 起始模型 | 起始工作量 | 原因 | ||
|---|---|---|---|---|---|
| 日常工程与调试。多 PR 功能开发 | Opus 5.5 | `medium`（默认），然后 `high` | 在日常任务中平衡成本与能力。 | ||
| 分类、路由、标记和高批量提取 | Haiku 5.5 | `medium`（默认） | 模型系列中成本最低且延迟最低。 | ||
| 搜索文件、读取日志或进行摘要的子代理 | Haiku 5.5 或 Sonnet 5.5 | 模型的默认值 | 错误答案易于发现，而且主代理无论如何都需要做出判断。 | ||
| 聊天和交互式会话 | Sonnet 5.5（或在大量简单聊天时使用 Haiku 5.5） | `low` 或 `medium` | Sonnet 5.5 速度很快，尤其是在 `low` 或 `medium` 工作量下。 | ||
| 范围明确的编码工作，如修复漏洞、功能迭代和规范驱动的工作 | Sonnet 5.5 | `medium`，对于更难的任务则为 `high` | 它能以 Opus 5.5 一半的 token 价格很好地处理范围明确的任务。 | ||
| 日常文档、幻灯片和电子表格 | Sonnet 5.5 | `high`（默认；在 Claude Code 中为 `medium`） | 它迭代速度快且具有出色的设计眼光。对于复杂的财务模型，或由大量源材料构建的文档，请从 Opus 5.5 开始。 | ||
| 你反复运行的代理任务，如调查、审查或起草 | Sonnet 5.5 | `medium` | 定义明确且重复的任务很少需要更大的模型。 | ||
| 计算机使用和视觉密集型工作流 | Opus 5.5 | `medium` | 这些运行涉及对许多截图和图像的长时间代理循环，这正是 Opus 5.5 所擅长的长时间运行的代理工作。 | ||
| 重大重构和迁移、架构规划以及领导多个子代理 | Fable 5.2 | `medium`，然后 `high`（API 默认为 `high`） | 它是 Anthropic 在技术领域最强大的通用可用模型。 | ||
| 深度研究，以及科学或定量分析 | Fable 5.2 | `medium`，然后 `high`（API 默认为 `high`） | 它是专为编码、金融和科学领域中最困难的技术项目而构建的。 | ||
| 你还不确定 | Opus 5.5 | `medium`（默认） | Claude Code 在 Pro、Max、Ultra、Team、Enterprise 和 Anthropic API 上默认使用它，平台文档也推荐其作为起点。 |
The four models side by side
四个模型并排对比
| Haiku 5.5 | Sonnet 5.5 | Opus 5.5 | Fable 5.2 | |||
|---|---|---|---|---|---|---|
| Built for | High-volume, cost-sensitive tasks | Fast, capable help with well-scoped tasks | Our daily driver for complex coding and knowledge work | The most ambitious, open-ended technical work | ||
| Relative latency | Fastest | Fast | Moderate | Moderate | ||
| Input / output, per million tokens | $0.10 / $0.50 for prompts up to 100K tokens, $0.50 / $2.50 over 100K | $2 / $10 | $4 / $20 | $8 / $40 | ||
| Cache reads, per million tokens | $0.01 up to 100K, $0.05 over 100K | $0.10 | $0.20 | $0.20 | ||
| Context window | 1M tokens | 1M tokens | 1M tokens | 1M tokens | ||
| Max output | 128K tokens | 128K tokens | 128K tokens | 128K tokens | ||
| Thinking on the API | Adaptive, on by default, and you can turn it off | Adaptive, on by default; `between_tools` skips upfront thinking | Adaptive, always on | Adaptive, always on | ||
| Default effort on the API (Claude Code can differ; `/effort status` shows yours) | `medium` | `high` (`medium` in Claude Code) | `medium` | `high` (`medium` in Claude Code) | ||
| Fast mode | No | No | Yes | Yes |
| Haiku 5.5 | Sonnet 5.5 | Opus 5.5 | Fable 5.2 | |||
|---|---|---|---|---|---|---|
| 适用场景 | 高吞吐量、对成本敏感的任务 | 针对范围明确的任务，提供快速且能力出众的帮助 | 我们日常用于复杂编码和知识工作的首选模型 | 最具野心、开放式的技术工作 | ||
| 相对延迟 | 最快 | 快速 | 中等 | 中等 | ||
| 每百万 token 的输入/输出费用 | 提示词不超过 100K tokens 时为 $0.10 / $0.50，超过 100K 时为 $0.50 / $2.50 | $2 / $10 | $4 / $20 | $8 / $40 | ||
| 每百万 token 的缓存读取费用 | 不超过 100K 时为 $0.01，超过 100K 时为 $0.05 | $0.10 | $0.20 | $0.20 | ||
| 上下文窗口 | 1M tokens | 1M tokens | 1M tokens | 1M tokens | ||
| 最大输出长度 | 128K tokens | 128K tokens | 128K tokens | 128K tokens | ||
| API 中的思考模式 | 自适应，默认开启，可关闭 | 自适应，默认开启；`between_tools` 跳过前置思考 | 自适应，始终开启 | 自适应，始终开启 | ||
| API 中的默认努力程度（Claude Code 可能不同；`/effort status` 可查看你的设置） | `medium` | `high`（在 Claude Code 中为 `medium`） | `medium` | `high`（在 Claude Code 中为 `medium`） | ||
| 快速模式 | 无 | 无 | 有 | 有 |
Prices are Claude API list prices. For benchmark results, see the Fable 5.2 page.
价格为 Claude API 列表价。基准测试结果请参阅 Fable 5.2 页面。
Video: Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5 each build the same portfolio site, shown side by side. Demo by Addy Osmani. — Four portfolio sites built by Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, shown side by side.
视频：Claude Fable 5.2、Opus 5.5、Sonnet 5.5 和 Haiku 5.5 各自构建同一个投资组合网站，并排展示。演示由 Addy Osmani 提供。—— 由 Fable 5.2、Opus 5.5、Sonnet 5.5 和 Haiku 5.5 构建的四个投资组合网站并排展示。
The trade-offs
权衡取舍
When you choose between models, you're making a trade-off. When you pick a larger model, for example, you are choosing capability over latency. These are the main trade-offs to think through.
当你在模型之间做出选择时，你是在进行权衡。例如，当你选择一个更大的模型时，你是在以延迟换取能力。以下是需要主要考虑的权衡因素。
Capability and cost per finished task
完成每项任务的能力与成本
Generally, bigger models cost more per token and are more capable, but the cost vs capability trade-off isn't universally true at the task level.
通常来说，较大的模型每个 token 的成本更高，能力也更强，但在任务层面，成本与能力的权衡并非绝对成立。
A higher price per token doesn't always mean a higher cost per task, or a better result. As we showed in Spending your effort, on Anthropic's runs of Terminal-Bench 3.0, the previous Fable, 5.1, at `max` scored 58.0%, while Opus 5.5 at `high` scored 58.9% with half the tokens.
每个 token 的价格较高并不总是意味着每项任务的总成本更高或结果更好。正如我们在“如何有效投入精力”一文中所展示的，在 Anthropic 运行 Terminal-Bench 3.0 的结果中，前代模型 Fable 5.1 在 `max` 设置下得分 58.0%，而 Opus 5.5 在 `high` 设置下仅用一半的 token 数就达到了 58.9% 的得分。
On the hardest work, the price gap can close. On FrontierCode 2.0, Fable 5.2 at `high` scored about level with Opus 5.5 at `max` (45.3% vs. 45.5%) for roughly half the cost per task. So before you raise Opus 5.5 past `high` on a hard task, test Fable 5.2. For everyday work, Opus 5.5 at `medium` is still the place to start. For benchmark results, see the launch post.
在最困难的工作中，价格差距可能会缩小。在 FrontierCode 2.0 上，Fable 5.2 在 `high` 设置下的得分约为 45.3%，与 Opus 5.5 在 `max` 设置下的 45.5% 基本持平，但每项任务的成本大约只有一半。因此，在对困难任务将 Opus 5.5 提升至 `high` 之前，请先测试 Fable 5.2。对于日常工作，Opus 5.5 在 `medium` 设置下仍是首选起点。基准测试结果请参阅发布博文。
However, the Claude Platform's cost and intelligence measurements also show the limits of a small model. These generally do well on high-volume work with checkable outputs and are less suited to long agentic loops.
然而，Claude 平台的成本与智能度指标也揭示了小模型的局限。这类模型通常在可验证输出的高吞吐量任务中表现良好，但不太适合长周期的智能体循环。
When you compare two models, price the hardest tasks in your set as well as the typical ones, because the hardest tasks carry most of the spend. On one 20-problem search benchmark, two problems carried 43% of the spend.
在比较两个模型时，应同时评估你数据集中最困难的典型任务，因为最困难的任务占据了大部分支出。在一个包含 20 个问题的搜索基准测试中，仅有两个问题就占用了 43% 的支出。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力