Claude Opus 5vsOpus 4.7
在 26 个同模式 benchmark 中，Claude Opus 5 整体领先：Claude Opus 5 领先 24 项，Opus 4.7 领先 1 项，持平 1 项，平均分差 +23.07。另有 12 项测试模式不同，仅供参考。
Claude Opus 5
Anthropic · 2026-07-24 · 推理大模型
Opus 4.7
Anthropic · 2026-04-16 · 推理大模型
Claude Opus 524 项(92%)持平1(4%)1 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 26 项。
数学
Claude Opus 5 领先 3/3| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| FrontierMath Tier 4 v2 | 73.1719 / 77Max (With Tools) | 31.7137 / 77Max (With Tools) | +41.46 |
| Riemann-bench | 684 / 42Max (With Tools + Internet) | 32.8021 / 42Max (With Tools + Internet) | +35.20 |
| FrontierMath v2 | 85.6112 / 120Max (With Tools) | 70.1827 / 120Max (With Tools) | +15.44 |
金融
Claude Opus 5 领先 3/3| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Tax Agent Bench | 45.5524 / 71Unspecified Reasoning (With Tools) | 23.7450 / 71Unspecified Reasoning (With Tools) | +21.81 |
| EMB (Excel Modeling) | 73.566 / 77Max (With Tools) | 63.7327 / 77Max (With Tools) | +9.83 |
| Finance Agent v2 | 58.6310 / 76Max (With Tools) | 51.5139 / 76Max (With Tools) | +7.12 |
临床工作与医疗决策
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| MedCode | 63.571 / 107Max (With Tools) | 54.867 / 107Max (With Tools) | +8.71 |
| MedScribe | 90.984 / 109Max (With Tools) | 82.9549 / 109Max (With Tools) | +8.03 |
客服与业务流程
胶着 2/2| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| τ³-Banking | 48.716 / 170Max (With Tools) | 40.2131 / 170Max (With Tools) | +8.50 |
| SAGE | 49.4321 / 91Max (With Tools) | 56.101 / 91Max (With Tools) | -6.67 |
抽象归纳与泛化
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| ARC-AGI-3 | 30.1610 / 66Thinking High (No Tools) | 0.1853 / 66Thinking High (No Tools) | +29.98 |
| ARC-AGI-1 | 97.5013 / 249Thinking High (No Tools) | 93.5053 / 249Thinking High (No Tools) | +4 |
法律
Claude Opus 5 领先 1/2| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Legal Research Bench | 55.291 / 84Max (With Tools) | 38.4635 / 84Max (With Tools) | +16.83 |
| Harvey's Legal Agent Benchmark | 6.6721 / 51Max (With Tools) | 6.6721 / 51Max (With Tools) | 持平 |
专业数据分析
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| AA-AnalystAgent | 53.753 / 29Max (With Tools + Internet) | 43.7511 / 29Max (With Tools + Internet) | +10 |
仓库修复与多文件工程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| APEX-SWE | 63.703 / 72Max (With Tools) | 47.6020 / 72Max (With Tools) | +16.10 |
写作与创意表达
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Creative Writing | 2,1333 / 114Normal (No Tools) | 1,91412 / 114Normal (No Tools) | +218.30 |
办公与文档流程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| AutomationBench | 50.3013 / 148Max (With Tools) | 13.3983 / 148Max (With Tools) | +36.91 |
图像感知与视觉推理
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| MMMU-Pro | 84.708 / 233Max (No Tools) | 78.8057 / 233Max (No Tools) | +5.90 |
多轮记忆与持续上下文
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| EBR-bench | 45.715 / 23Max (With Tools) | 19.0513 / 23Max (With Tools) | +26.66 |
机器学习工程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| WeirdML v2 | 86.344 / 52Normal (With Tools) | 76.4013 / 52Normal (With Tools) | +9.94 |
科学知识与推理
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| CritPt | 29.1012 / 208Max (No Tools) | 1271 / 208Max (No Tools) | +17.10 |
程序生成与编辑
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 88.4010 / 114Max (With Tools) | 7141 / 114Max (With Tools) | +17.40 |
维护、优化与工程管理
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Code Migration | 57.478 / 79Max (With Tools) | 43.8826 / 79Max (With Tools) | +13.59 |
跨应用与工具编排
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| APEX-Agents | 65.806 / 61Max (With Tools) | 49.2030 / 61Max (With Tools) | +16.60 |
跨能力聚合指数
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.7 | 分差 |
|---|---|---|---|
| Vals Index | 67.214 / 42Max (With Tools) | 56.1124 / 42Max (With Tools) | +11.10 |
测试模式不同的成绩
共 12 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 Opus 5 | Opus 4.7 |
|---|---|---|
| FrontierCode 1.1 Main | 53.40Reported best (effort unspecified) | 38.50Max (With Tools) |
| SimpleBench | 80.60Thinking High (No Tools) | 61.70Normal (No Tools) |
| LiveBench | 80.08Max (No Tools) | 76.53Deep Thinking (No Tools) |
| BrowseComp | 90.80Max (With Tools + Internet) | 79.30Extended (With Tools) |
| HLE | 63.60Max (With Tools) | 54.70Extended (With Tools) |
| Context Arena | 97.72Max (No Tools) | 22.12Normal (No Tools) |
| PostTrain Bench | 35Reported best (effort unspecified) | 28.60Extra-High (With Tools) |
| Text Arena (Coding) | 1,712Max (No Tools) | 1,562Normal (No Tools) |
| SWE-Bench Pro - Public | 79.20Max (With Tools) | 64.30Extended (With Tools) |
| SWE-bench Verified | 96Max (With Tools) | 87.60Extended (With Tools) |
| GPQA Diamond | 87.88Thinking Low (No Tools) | 94.20Extended (No Tools) |
| MCP-Atlas | 85.80Extra-High (With Tools) | 79.10Deep Thinking (With Tools) |
规格对比
| 字段 | Claude Opus 5 | Opus 4.7 |
|---|---|---|
| 发布机构 | Anthropic | Anthropic |
| 发布时间 | 2026-07-24 | 2026-04-16 |
| 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 | 1M | 1000K |
| 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 Opus 5 | Opus 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 5在以下类目领先:数学 (3/3)、金融 (3/3)、临床工作与医疗决策 (2/2)、抽象归纳与泛化 (2/2)、法律 (1/2)、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)、办公与文档流程 (1/1)、图像感知与视觉推理 (1/1)、多轮记忆与持续上下文 (1/1)、机器学习工程 (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)、跨应用与工具编排 (1/1)、跨能力聚合指数 (1/1)
- 胶着类目:客服与业务流程
26 个同模式 benchmark 上，Claude Opus 5 平均高出 23.07 分。
单项差距最大的 benchmark：Creative Writing — Claude Opus 5 2,133，Opus 4.7 1,914（分差 +218.30）。
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。