Claude Opus 5vsOpus 4.5
在 3 个同模式 benchmark 中，Claude Opus 5 整体领先：Claude Opus 5 领先 3 项，Opus 4.5 领先 0 项，持平 0 项，平均分差 +174.22。另有 21 项测试模式不同，仅供参考。
Claude Opus 5
Anthropic · 2026-07-24 · 推理大模型
Opus 4.5
Anthropic · 2025-11-25 · 推理大模型
Claude Opus 53 项(100%)(0%)0 项Opus 4.5
评测分数
按能力类目分组，每组内按分差大小排列；共 3 项。
写作与创意表达
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.5 | 分差 |
|---|---|---|---|
| Creative Writing | 2,1333 / 114Normal (No Tools) | 1,68732 / 114Normal (No Tools) | +445.90 |
客服与业务流程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.5 | 分差 |
|---|---|---|---|
| τ³-Banking | 44.7018 / 170Thinking High (With Tools) | 24.7481 / 170Thinking High (With Tools) | +19.96 |
数学
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | Opus 4.5 | 分差 |
|---|---|---|---|
| Riemann-bench | 684 / 42Max (With Tools + Internet) | 11.2036 / 42Max (With Tools + Internet) | +56.80 |
测试模式不同的成绩
共 21 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 4.5 |
|---|---|---|
| ARC-AGI-1 | 97.50Thinking High (No Tools) | 40Normal (No Tools) |
| MedCode | 63.57Max (With Tools) | 45.17Normal (With Tools) |
| MedScribe | 90.98Max (With Tools) | 83.25Normal (With Tools) |
| Vibe Code Bench v1.1 | 88.40Max (With Tools) | 20.63Thinking High (With Tools) |
| SimpleBench | 80.60Thinking High (No Tools) | 62Extended (No Tools) |
| LiveBench | 80.08Max (No Tools) | 72.58Thinking (No Tools, 64K Budget) |
| HLE | 63.60Max (With Tools) | 14.16Normal (No Tools) |
| BALROG | 63.40Max (With Tools) | 43.50Normal (With Tools) |
| FrontierMath Tier 4 v2 | 73.17Max (With Tools) | 4.88Thinking (With Tools, 32K Budget) |
| FrontierMath v2 | 85.61Max (With Tools) | 34.39Thinking (With Tools, 32K Budget) |
| EBR-bench | 45.71Max (With Tools) | 14.29Thinking (With Tools, 128K Budget) |
| WeirdML v2 | 86.34Normal (With Tools) | 63.70Thinking (With Tools, 16K Budget) |
| WeirdML v3 | 0.19Extra-High (With Tools) | 0.07Thinking High (With Tools) |
| Text Arena (Coding) | 1,712Max (No Tools) | 1,479Normal (No Tools) |
| APEX-SWE | 63.70Max (With Tools) | 38.70Thinking High (With Tools) |
| SWE-bench Verified | 96Max (With Tools) | 80.90Extended (With Tools) |
| CritPt | 29.10Max (No Tools) | 0.30Normal (No Tools) |
| GPQA Diamond | 87.88Thinking Low (No Tools) | 87Extended (No Tools) |
| SAGE | 49.43Max (With Tools) | 45Normal (With Tools) |
| MCP-Atlas | 85.80Extra-High (With Tools) | 69.80Thinking High (With Tools) |
| MMMU-Pro | 84.70Max (No Tools) | 71.20Normal (No Tools) |
规格对比
| 字段 | Claude Opus 5 | Opus 4.5 |
|---|---|---|
| 发布机构 | Anthropic | Anthropic |
| 发布时间 | 2026-07-24 | 2025-11-25 |
| 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 | 200K |
| 最大输出 | 128K | 64K |
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 4.5 |
|---|---|---|
| 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:写作与创意表达 (1/1)、客服与业务流程 (1/1)、数学 (1/1)
3 个同模式 benchmark 上，Claude Opus 5 平均高出 174.22 分。
单项差距最大的 benchmark：Creative Writing — Claude Opus 5 2,133，Opus 4.5 1,687（分差 +445.90）。
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。