DataLearner 标志

Claude Fable 5vsClaude Opus 4.8

在 30 个共同 benchmark 中,Claude Fable 5 整体领先:Claude Fable 5 领先 27 项,Claude Opus 4.8 领先 2 项,持平 1 项,平均分差 +8.68。

Anthropic
Claude Fable 5

Anthropic · 2026-06-09 · 推理大模型

Anthropic
Claude Opus 4.8

Anthropic · 2026-05-28 · 推理大模型

Claude Fable 527 项(90%)持平1(7%)2 项Claude Opus 4.8

评测分数

按能力类目分组,每组内按分差大小排列;共 30 项。

客服与业务流程

胶着 3/3
评测项Claude Fable 5Claude Opus 4.8分差
SAGE51.897 / 64Max (With Tools)54.792 / 64Max (With Tools)-2.90
Public Benefits Bench v1.170.434 / 26Max (With Tools)68.138 / 26Max (With Tools)+2.30
τ³-Banking39.6934 / 167Max (With Tools)39.6934 / 167Max (With Tools)持平

法律

Claude Fable 5 领先 3/3
评测项Claude Fable 5Claude Opus 4.8分差
Legal Research Bench49.525 / 42Max (With Tools)43.7512 / 42Max (With Tools)+5.77
Harvey Lab-AA93.563 / 44Max (With Tools)91.0811 / 44Max (With Tools)+2.48
Harvey's Legal Agent Benchmark11.258 / 43Max (With Tools)9.5812 / 43Max (With Tools)+1.67

临床工作与医疗决策

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
MedCode56.073 / 64Max (With Tools)53.228 / 64Max (With Tools)+2.85
MedScribe88.529 / 66Max (With Tools)85.7517 / 66Max (With Tools)+2.77

数学

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
FrontierMath Tier 4 v287.802 / 42Max (No Tools)56.1011 / 42Max (No Tools)+31.71
FrontierMath v287.023 / 58Max (No Tools)809 / 58Max (No Tools)+7.02

科学计算与科研编程

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
Terminal-Bench-Science 0.121.407 / 18Max (With Tools)10.5011 / 18Max (With Tools)+10.90
SciCode613 / 134Max (No Tools)54.4044 / 134Max (No Tools)+6.60

自主开发与终端任务

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
Terminal-Bench 4.044.5518 / 98Max (With Tools)23.6435 / 98Max (With Tools)+20.91
Terminal Bench Hard62.902 / 244Max (With Tools)58.3010 / 244Max (With Tools)+4.60

跨应用与工具编排

胶着 2/2
评测项Claude Fable 5Claude Opus 4.8分差
PinchBench v254.8435 / 45Reported best (effort unspecified)90.542 / 45Reported best (effort unspecified)-35.70
MCPMark-Verified86.614 / 8Max (With Tools)76.386 / 8Max (With Tools)+10.23

跨能力聚合指数

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
Vals Index v2.161.397 / 36Max (task-specific tools)55.1012 / 36Max (task-specific tools)+6.29
Vals Index66.046 / 42Max (With Tools)60.9111 / 42Max (With Tools)+5.13

金融

Claude Fable 5 领先 2/2
评测项Claude Fable 5Claude Opus 4.8分差
EMB (Excel Modeling)73.673 / 42Max (With Tools)69.3710 / 42Max (With Tools)+4.30
Finance Agent v256.3113 / 43Max (With Tools)53.9219 / 43Max (With Tools)+2.39

专业数据分析

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
AA-AnalystAgent48.756 / 29Max (With Tools + Internet)459 / 29Max (With Tools + Internet)+3.75

仓库修复与多文件工程

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
DeepSWE69.9118 / 92Deep Thinking (With Tools)58.9751 / 92Deep Thinking (With Tools)+10.94

写作与创意表达

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
Creative Writing1,94310 / 110Normal (No Tools)1,84018 / 110Normal (No Tools)+103.30

多轮记忆与持续上下文

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
EBR-bench39.527 / 23Max (With Tools)28.5710 / 23Max (With Tools)+10.95

常识推理

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
SimpleBench81.906 / 96Normal (No Tools)64.8024 / 96Normal (No Tools)+17.10

文档与图表理解

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
GDP.pdf2421 / 122Max (No Tools)22.8026 / 122Max (No Tools)+1.20

科学知识与推理

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
CritPt28.6016 / 204Max (No Tools)20.9035 / 204Max (No Tools)+7.70

程序生成与编辑

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
Vibe Code Bench v1.190.352 / 61Max (With Tools)82.7213 / 61Max (With Tools)+7.63

维护、优化与工程管理

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
Code Migration55.065 / 43Max (With Tools)47.2510 / 43Max (With Tools)+7.81

跨能力综合测试

Claude Fable 5 领先 1/1
评测项Claude Fable 5Claude Opus 4.8分差
LiveBench79.532 / 117Deep Thinking (No Tools)78.933 / 117Deep Thinking (No Tools)+0.60

规格对比

字段Claude Fable 5Claude Opus 4.8
发布机构AnthropicAnthropic
发布时间2026-06-092026-05-28
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1M1M
最大输出128K125K

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 5Claude Opus 4.8
文本输入$10 / 1M tokens$5 / 1M tokens
文本输出$50 / 1M tokens$25 / 1M tokens
缓存读取$1 / 1M tokens$0.5 / 1M tokens
缓存写入$12.5 / 1M tokens$6.25 / 1M tokens

小结

  • Claude Fable 5在以下类目领先:法律 (3/3)、临床工作与医疗决策 (2/2)、数学 (2/2)、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)、仓库修复与多文件工程 (1/1)、写作与创意表达 (1/1)、多轮记忆与持续上下文 (1/1)、常识推理 (1/1)、文档与图表理解 (1/1)、科学知识与推理 (1/1)、程序生成与编辑 (1/1)、维护、优化与工程管理 (1/1)、跨能力综合测试 (1/1)
  • 胶着类目:客服与业务流程、跨应用与工具编排

30 个共同 benchmark 上,Claude Fable 5 平均高出 8.68 分。

单项差距最大的 benchmark:Creative Writing — Claude Fable 5 1,943,Claude Opus 4.8 1,840(分差 +103.30)。

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。