Claude Opus 5vsGLM-5.2
在 27 个同模式 benchmark 中，Claude Opus 5 整体领先：Claude Opus 5 领先 25 项，GLM-5.2 领先 1 项，持平 1 项，平均分差 +38.23。另有 10 项测试模式不同，仅供参考。
Claude Opus 5
Anthropic · 2026-07-24 · 推理大模型
GLM-5.2
智谱AI · 2026-06-13 · 推理大模型
Claude Opus 525 项(93%)持平1(4%)1 项GLM-5.2
评测分数
按能力类目分组，每组内按分差大小排列；共 27 项。
数学
Claude Opus 5 领先 3/3| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Riemann-bench | 684 / 42Max (With Tools + Internet) | 10.4038 / 42Max (With Tools + Internet) | +57.60 |
| FrontierMath Tier 4 v2 | 73.1719 / 77Max (With Tools) | 29.2741 / 77Max (With Tools) | +43.90 |
| FrontierMath v2 | 85.6112 / 120Max (With Tools) | 59.2141 / 120Max (With Tools) | +26.41 |
法律
Claude Opus 5 领先 2/3| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Legal Research Bench | 55.291 / 84Max (With Tools) | 31.2547 / 84Max (With Tools) | +24.04 |
| Harvey Lab-AA | 93.464 / 44Max (With Tools) | 90.9712 / 44Max (With Tools) | +2.49 |
| Harvey's Legal Agent Benchmark | 6.6721 / 51Max (With Tools) | 7.0820 / 51Max (With Tools) | -0.41 |
临床工作与医疗决策
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| MedCode | 63.571 / 107Max (With Tools) | 40.7760 / 107Max (With Tools) | +22.80 |
| MedScribe | 90.984 / 109Max (With Tools) | 83.5346 / 109Max (With Tools) | +7.45 |
仓库修复与多文件工程
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| DeepSWE | 73.6512 / 106Max (With Tools) | 43.7887 / 106Max (With Tools) | +29.87 |
| APEX-SWE | 63.703 / 72Max (With Tools) | 37.4045 / 72Max (With Tools) | +26.30 |
多轮记忆与持续上下文
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| EBR-bench | 45.715 / 23Max (With Tools) | 9.5218 / 23Max (With Tools) | +36.19 |
| Context Arena | 97.721 / 154Max (No Tools) | 72.3466 / 154Max (No Tools) | +25.38 |
科学知识与推理
Claude Opus 5 领先 1/2| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| CritPt | 29.1012 / 208Max (No Tools) | 20.9037 / 208Max (No Tools) | +8.20 |
| GPQA Diamond | 87.8865 / 250Thinking Low (No Tools) | 87.8865 / 250Thinking Low (No Tools) | 持平 |
金融
Claude Opus 5 领先 2/2| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| EMB (Excel Modeling) | 73.566 / 77Max (With Tools) | 61.5334 / 77Max (With Tools) | +12.03 |
| Finance Agent v2 | 58.6310 / 76Max (With Tools) | 49.7045 / 76Max (With Tools) | +8.93 |
写作与创意表达
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Creative Writing | 2,1333 / 114Normal (No Tools) | 1,75725 / 114Normal (No Tools) | +376.10 |
办公与文档流程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| AutomationBench | 50.3013 / 148Max (With Tools) | 7.76110 / 148Max (With Tools) | +42.54 |
客服与业务流程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| τ³-Banking | 48.716 / 170Max (With Tools) | 34.6050 / 170Max (With Tools) | +14.11 |
文档与图表理解
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| GDP.pdf | 21.6034 / 126Max (No Tools) | 10.4086 / 126Max (No Tools) | +11.20 |
机器学习工程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| WeirdML v2 | 91.591 / 52Thinking High (With Tools) | 67.3121 / 52Thinking High (With Tools) | +24.28 |
科学计算与科研编程
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| SciCode | 56.4028 / 137Max (No Tools) | 51.2070 / 137Max (No Tools) | +5.20 |
程序生成与编辑
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 88.4010 / 114Max (With Tools) | 63.9650 / 114Max (With Tools) | +24.44 |
维护、优化与工程管理
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Code Migration | 57.478 / 79Max (With Tools) | 37.8734 / 79Max (With Tools) | +19.60 |
综合偏好评测
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Text Arena (Coding) | 1,7121 / 35Max (No Tools) | 1,5935 / 35Max (No Tools) | +118.63 |
自主开发与终端任务
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 51.8217 / 108Max (With Tools) | 196 / 108Max (With Tools) | +50.82 |
跨能力聚合指数
Claude Opus 5 领先 1/1| 评测项 | Claude Opus 5 | GLM-5.2 | 分差 |
|---|---|---|---|
| Vals Index | 67.214 / 42Max (With Tools) | 53.1227 / 42Max (With Tools) | +14.09 |
测试模式不同的成绩
共 10 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 | GLM-5.2 |
|---|---|---|
| FrontierCode 1.1 Main | 53.40Reported best (effort unspecified) | 24.50Normal (With Tools) |
| Program Bench | 37Max (With Tools) | 63.70Thinking (With Tools) |
| SimpleBench | 80.60Thinking High (No Tools) | 58.80Normal (No Tools) |
| LiveBench | 80.08Max (No Tools) | 73.18Normal (No Tools) |
| HLE | 63.60Max (With Tools) | 54.70Thinking (With Tools) |
| PostTrain Bench | 35Reported best (effort unspecified) | 34.30Max (With Tools) |
| NL2Repo-Bench | 75.30Max (With Tools) | 48.90Thinking (With Tools) |
| SWE-Bench Pro - Public | 79.20Max (With Tools) | 62.10Thinking (With Tools) |
| APEX-Agents | 65.80Max (With Tools) | 45.20Reported best (effort unspecified) |
| MCP-Atlas | 85.80Extra-High (With Tools) | 77.80Thinking (With Tools) |
规格对比
| 字段 | Claude Opus 5 | GLM-5.2 |
|---|---|---|
| 发布机构 | Anthropic | 智谱AI |
| 发布时间 | 2026-07-24 | 2026-06-13 |
| 模型类型 | 推理大模型 | 推理大模型 |
| 架构 | 稠密模型 | MoE 架构 |
| 参数规模 | 暂无数据 | 7533.3亿 |
| 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 Opus 5 | GLM-5.2 |
|---|---|---|
| 文本输入 | $5 / 1M tokens | ¥8 / 1M tokens |
| 文本输出 | $25 / 1M tokens | ¥28 / 1M tokens |
| 缓存读取 | $0.5 / 1M tokens | ¥2 / 1M tokens |
| 缓存写入 | $6.25 / 1M tokens | 暂无公开价格 |
小结
- Claude Opus 5在以下类目领先:数学 (3/3)、法律 (2/3)、临床工作与医疗决策 (2/2)、仓库修复与多文件工程 (2/2)、多轮记忆与持续上下文 (2/2)、科学知识与推理 (1/2)、金融 (2/2)、写作与创意表达 (1/1)、办公与文档流程 (1/1)、客服与业务流程 (1/1)、文档与图表理解 (1/1)、机器学习工程 (1/1)、科学计算与科研编程 (1/1)、程序生成与编辑 (1/1)、维护、优化与工程管理 (1/1)、综合偏好评测 (1/1)、自主开发与终端任务 (1/1)、跨能力聚合指数 (1/1)
27 个同模式 benchmark 上，Claude Opus 5 平均高出 38.23 分。
单项差距最大的 benchmark：Creative Writing — Claude Opus 5 2,133，GLM-5.2 1,757（分差 +376.10）。
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。