DataLearner 标志

Claude Opus 4.8vsOpus 4.7

在 28 个共同 benchmark 中,Claude Opus 4.8 整体领先:Claude Opus 4.8 领先 22 项,Opus 4.7 领先 6 项,持平 0 项,平均分差 +1.10。

Anthropic
Claude Opus 4.8

Anthropic · 2026-05-28 · 推理大模型

Anthropic
Opus 4.7

Anthropic · 2026-04-16 · 推理大模型

Claude Opus 4.822 项(79%)(21%)6 项Opus 4.7

评测分数

按能力类目分组,每组内按分差大小排列;共 28 项。

临床工作与医疗决策

胶着 2/2
评测项Claude Opus 4.8Opus 4.7分差
MedScribe85.7517 / 66Max (With Tools)82.9534 / 66Max (With Tools)+2.80
MedCode53.228 / 64Max (With Tools)54.865 / 64Max (With Tools)-1.64

仓库修复与多文件工程

Claude Opus 4.8 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
SWE-Bench Pro - Public69.205 / 62Extended (With Tools)64.3010 / 62Extended (With Tools)+4.90
SWE-bench Verified88.605 / 116Extended (With Tools)87.606 / 116Extended (With Tools)+1

客服与业务流程

Opus 4.7 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
SAGE54.792 / 64Max (With Tools)56.101 / 64Max (With Tools)-1.31
τ³-Banking39.6934 / 167Max (With Tools)40.2131 / 167Max (With Tools)-0.52

数学

Claude Opus 4.8 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
FrontierMath Tier 4 v256.1011 / 42Max (No Tools)31.7117 / 42Max (No Tools)+24.39
FrontierMath v2809 / 58Max (No Tools)70.1815 / 58Max (No Tools)+9.82

法律

Claude Opus 4.8 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
Legal Research Bench43.7512 / 42Max (With Tools)38.4622 / 42Max (With Tools)+5.29
Harvey's Legal Agent Benchmark9.5812 / 43Max (With Tools)6.6717 / 43Max (With Tools)+2.91

跨应用与工具编排

Claude Opus 4.8 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
PinchBench v290.542 / 45Reported best (effort unspecified)76.0115 / 45Reported best (effort unspecified)+14.53
MCP-Atlas82.2010 / 44Deep Thinking (With Tools)79.1014 / 44Deep Thinking (With Tools)+3.10

金融

Claude Opus 4.8 领先 2/2
评测项Claude Opus 4.8Opus 4.7分差
EMB (Excel Modeling)69.3710 / 42Max (With Tools)63.7320 / 42Max (With Tools)+5.64
Finance Agent v253.9219 / 43Max (With Tools)51.5127 / 43Max (With Tools)+2.41

专业数据分析

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
AA-AnalystAgent459 / 29Max (With Tools + Internet)43.7511 / 29Max (With Tools + Internet)+1.25

写作与创意表达

Opus 4.7 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Creative Writing1,84018 / 110Normal (No Tools)1,91412 / 110Normal (No Tools)-74.50

多轮记忆与持续上下文

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
EBR-bench28.5710 / 23Max (With Tools)19.0513 / 23Max (With Tools)+9.52

常识推理

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
SimpleBench64.8024 / 96Normal (No Tools)61.7030 / 96Normal (No Tools)+3.10

抽象归纳与泛化

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
ARC-AGI-31.5222 / 51Thinking High (No Tools)0.1838 / 51Thinking High (No Tools)+1.34

机器学习工程

Opus 4.7 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
WeirdML v270.4518 / 52Normal (With Tools)76.4013 / 52Normal (With Tools)-5.95

桌面工作流

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
OSWorld-Verified83.404 / 28Extended (With Tools)7813 / 28Extended (With Tools)+5.40

程序生成与编辑

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Vibe Code Bench v1.182.7213 / 61Max (With Tools)7125 / 61Max (With Tools)+11.72

维护、优化与工程管理

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Code Migration47.2510 / 43Max (With Tools)43.8818 / 43Max (With Tools)+3.37

综合偏好评测

Opus 4.7 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Text Arena (Coding)1,5459 / 35Normal (No Tools)1,5627 / 35Normal (No Tools)-17.34

综合知识考试

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
HLE57.9015 / 235Extended (With Tools)54.7026 / 235Extended (With Tools)+3.20

自主开发与终端任务

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Terminal-Bench 2.178.8831 / 57Thinking High (With Tools)69.7042 / 57Thinking High (With Tools)+9.18

跨能力综合测试

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
LiveBench78.933 / 117Deep Thinking (No Tools)76.538 / 117Deep Thinking (No Tools)+2.40

跨能力聚合指数

Claude Opus 4.8 领先 1/1
评测项Claude Opus 4.8Opus 4.7分差
Vals Index60.9111 / 42Max (With Tools)56.1124 / 42Max (With Tools)+4.80

规格对比

字段Claude Opus 4.8Opus 4.7
发布机构AnthropicAnthropic
发布时间2026-05-282026-04-16
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1M1000K
最大输出125K128K

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 Opus 4.8Opus 4.7
文本输入$5 / 1M tokens$5 / 1M tokens
文本输出$25 / 1M tokens$25 / 1M tokens
缓存读取$0.5 / 1M tokens$0.5 / 1M tokens
缓存写入$6.25 / 1M tokens$6.25 / 1M tokens

小结

  • Claude Opus 4.8在以下类目领先:仓库修复与多文件工程 (2/2)、数学 (2/2)、法律 (2/2)、跨应用与工具编排 (2/2)、金融 (2/2)、专业数据分析 (1/1)、多轮记忆与持续上下文 (1/1)、常识推理 (1/1)、抽象归纳与泛化 (1/1)、桌面工作流 (1/1)、程序生成与编辑 (1/1)、维护、优化与工程管理 (1/1)、综合知识考试 (1/1)、自主开发与终端任务 (1/1)、跨能力综合测试 (1/1)、跨能力聚合指数 (1/1)
  • Opus 4.7在以下类目领先:客服与业务流程 (2/2)、写作与创意表达 (1/1)、机器学习工程 (1/1)、综合偏好评测 (1/1)
  • 胶着类目:临床工作与医疗决策

28 个共同 benchmark 上,Claude Opus 4.8 平均高出 1.10 分。

单项差距最大的 benchmark:Creative Writing — Claude Opus 4.8 1,840,Opus 4.7 1,914(分差 -74.50)。

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。