Claude Fable 5.1vsClaude Fable 5
在 32 个同模式 benchmark 中，Claude Fable 5.1 整体领先：Claude Fable 5.1 领先 23 项，Claude Fable 5 领先 9 项，持平 0 项，平均分差 +10.14。另有 6 项测试模式不同，仅供参考。
Claude Fable 5.1
Anthropic · 2026-09-01 · 推理大模型
Claude Fable 5
Anthropic · 2026-06-09 · 推理大模型
Claude Fable 5.123 项(72%)(28%)9 项Claude Fable 5
评测分数
按能力类目分组，每组内按分差大小排列；共 32 项。
仓库修复与多文件工程
Claude Fable 5.1 领先 2/3| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| FrontierSWE v2 | 56.294 / 22Max (With Tools) | 46.967 / 22Max (With Tools) | +9.33 |
| APEX-SWE | 63.604 / 73Max (With Tools) | 58.805 / 73Max (With Tools) | +4.80 |
| DeepSWE | 67.4035 / 100Max (With Tools) | 69.7223 / 100Max (With Tools) | -2.32 |
客服与业务流程
Claude Fable 5.1 领先 2/3| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| τ³-Banking | 47.2010 / 168Max (With Tools) | 39.6934 / 168Max (With Tools) | +7.51 |
| Public Benefits Bench v1.1 | 74.902 / 26Max (With Tools) | 70.434 / 26Max (With Tools) | +4.47 |
| SAGE | 48.5320 / 65Max (With Tools) | 51.897 / 65Max (With Tools) | -3.36 |
数学
Claude Fable 5.1 领先 2/3| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| ProofBench v1.1 (Lean 4) | 1001 / 29Max (With Tools) | 955 / 29Max (With Tools) | +5 |
| FrontierMath v2 | 90.184 / 114Max (With Tools) | 87.029 / 114Max (With Tools) | +3.16 |
| FrontierMath Tier 4 v2 | 87.8010 / 71Max (With Tools) | 90.208 / 71Max (With Tools) | -2.40 |
法律
Claude Fable 5 领先 2/3| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Legal Research Bench | 55.291 / 46Max (With Tools) | 49.526 / 46Max (With Tools) | +5.77 |
| Harvey's Legal Agent Benchmark | 6.6720 / 50Max (With Tools) | 11.2511 / 50Max (With Tools) | -4.58 |
| Harvey Lab-AA | 93.026 / 44Max (With Tools) | 93.563 / 44Max (With Tools) | -0.54 |
临床工作与医疗决策
胶着 2/2| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| MedScribe | 91.292 / 68Max (With Tools) | 88.529 / 68Max (With Tools) | +2.77 |
| MedCode | 53.516 / 66Max (With Tools) | 56.073 / 66Max (With Tools) | -2.56 |
科学计算与科研编程
Claude Fable 5.1 领先 2/2| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 404 / 18Max (With Tools) | 21.407 / 18Max (With Tools) | +18.60 |
| SciCode | 63.102 / 135Max (No Tools) | 613 / 135Max (No Tools) | +2.10 |
自主开发与终端任务
Claude Fable 5.1 领先 2/2| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 57.886 / 104Max (With Tools) | 44.5521 / 104Max (With Tools) | +13.33 |
| CursorBench 3.2 | 73.401 / 5Max (With Tools) | 70.502 / 5Max (With Tools) | +2.90 |
跨能力聚合指数
Claude Fable 5.1 领先 2/2| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Vals Index v2.1 | 65.834 / 37Max (task-specific tools) | 61.397 / 37Max (task-specific tools) | +4.44 |
| Vals Index | 68.833 / 42Max (With Tools) | 66.046 / 42Max (With Tools) | +2.79 |
金融
Claude Fable 5.1 领先 2/2| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| EMB (Excel Modeling) | 76.671 / 46Max (With Tools) | 73.675 / 46Max (With Tools) | +3 |
| Finance Agent v2 | 58.888 / 46Max (With Tools) | 56.3115 / 46Max (With Tools) | +2.57 |
专业数据分析
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| AA-AnalystAgent | 57.502 / 29Max (With Tools + Internet) | 48.756 / 29Max (With Tools + Internet) | +8.75 |
写作与创意表达
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Creative Writing | 2,1622 / 112Normal (No Tools) | 1,94310 / 112Normal (No Tools) | +218.90 |
多轮记忆与持续上下文
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| EBR-bench | 57.143 / 23Max (With Tools) | 39.527 / 23Max (With Tools) | +17.62 |
抽象归纳与泛化
Claude Fable 5 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| ARC-AGI-1 | 97.5011 / 226Max (No Tools) | 98.501 / 226Max (No Tools) | -1 |
文档与图表理解
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| GDP.pdf | 26.2016 / 124Max (No Tools) | 2423 / 124Max (No Tools) | +2.20 |
机器学习工程
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| WeirdML v3 | 0.264 / 21Extra-High (With Tools) | 0.168 / 21Extra-High (With Tools) | +0.10 |
科学知识与推理
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| CritPt | 29.7011 / 204Max (No Tools) | 28.6016 / 204Max (No Tools) | +1.10 |
程序生成与编辑
Claude Fable 5 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 90.265 / 66Max (With Tools) | 90.353 / 66Max (With Tools) | -0.09 |
维护、优化与工程管理
Claude Fable 5 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| Code Migration | 54.619 / 48Max (With Tools) | 55.068 / 48Max (With Tools) | -0.45 |
跨能力综合测试
Claude Fable 5.1 领先 1/1| 评测项 | Claude Fable 5.1 | Claude Fable 5 | 分差 |
|---|---|---|---|
| LiveBench | 83.411 / 157Max (No Tools) | 82.973 / 157Max (No Tools) | +0.44 |
测试模式不同的成绩
共 6 项。两款模型的公开成绩来自不同测试模式，分数不直接可比，不计入胜负和平均分差。
| 评测项 | Claude Fable 5.1 | Claude Fable 5 |
|---|---|---|
| FrontierCode 1.1 Extended | 63.60Thinking Medium (With Tools) | 64.90Extra-High (With Tools) |
| FrontierCode 1.1 Main | 50.90Thinking Medium (With Tools) | 53.50Extra-High (With Tools) |
| SimpleBench | 86.60Thinking (No Tools) | 81.90Normal (No Tools) |
| HLE | 65Max (With Tools) | 59Deep Thinking (No Tools) |
| SWE-Bench Pro - Public | 81.20Max (With Tools) | 80.30Deep Thinking (With Tools) |
| MCP-Atlas | 87.20Max (With Tools) | 83.30Normal (With Tools) |
规格对比
| 字段 | Claude Fable 5.1 | Claude Fable 5 |
|---|---|---|
| 发布机构 | Anthropic | Anthropic |
| 发布时间 | 2026-09-01 | 2026-06-09 |
| 模型类型 | 推理大模型 | 推理大模型 |
| 架构 | 稠密模型 | 稠密模型 |
| 参数规模 | 暂无数据 | 暂无数据 |
| 上下文长度 | 1M | 1M |
| 最大输出 | 128K | 128K |
API 调用价格
价格优先使用 DataLearner 配置的 API 记录；缺失项不做推测。
| 价格项 | Claude Fable 5.1 | Claude Fable 5 |
|---|---|---|
| 文本输入 | $10 / 1M tokens | $10 / 1M tokens |
| 文本输出 | $50 / 1M tokens | $50 / 1M tokens |
| 缓存读取 | $0.25 / 1M tokens | $1 / 1M tokens |
| 缓存写入 | $12.5 / 1M tokens | $12.5 / 1M tokens |
小结
- Claude Fable 5.1在以下类目领先:仓库修复与多文件工程 (2/3)、客服与业务流程 (2/3)、数学 (2/3)、科学计算与科研编程 (2/2)、自主开发与终端任务 (2/2)、跨能力聚合指数 (2/2)、金融 (2/2)、专业数据分析 (1/1)、写作与创意表达 (1/1)、多轮记忆与持续上下文 (1/1)、文档与图表理解 (1/1)、机器学习工程 (1/1)、科学知识与推理 (1/1)、跨能力综合测试 (1/1)
- Claude Fable 5在以下类目领先:法律 (2/3)、抽象归纳与泛化 (1/1)、程序生成与编辑 (1/1)、维护、优化与工程管理 (1/1)
- 胶着类目:临床工作与医疗决策
32 个同模式 benchmark 上，Claude Fable 5.1 平均高出 10.14 分。
单项差距最大的 benchmark：Creative Writing — Claude Fable 5.1 2,162，Claude Fable 5 1,943（分差 +218.90）。
本页正文由结构化模型、价格与 benchmark 数据生成，不使用实时 LLM 撰写。