GLM-5.3vsClaude Opus 5
在 31 个共同 benchmark 中，Claude Opus 5 整体领先：GLM-5.3 领先 3 项，Claude Opus 5 领先 28 项，持平 0 项，平均分差 -13.01。
GLM-5.3
智谱AI · 2026-08-14 · 推理大模型
Claude Opus 5
Anthropic · 2026-07-24 · 推理大模型
GLM-5.33 项(10%)(90%)28 项Claude Opus 5
评测分数
按能力类目分组，每组内按分差大小排列；共 31 项。
数学
Claude Opus 5 领先 3/3| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| ProofBench v1.1 (Lean 4) | 4919 / 28Max (With Tools) | 993 / 28Max (With Tools) | -50 |
| FrontierMath Tier 4 v2 | 29.2720 / 42Max (No Tools) | 73.176 / 42Max (No Tools) | -43.90 |
| FrontierMath v2 | 68.7716 / 58Max (No Tools) | 85.615 / 58Max (No Tools) | -16.84 |
金融
Claude Opus 5 领先 3/3| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| EMB (Excel Modeling) | 56.3430 / 42Max (With Tools) | 73.564 / 42Max (With Tools) | -17.22 |
| Finance Agent v2 | 55.8415 / 43Max (With Tools) | 58.638 / 43Max (With Tools) | -2.79 |
| Tax Agent Bench | 73.093 / 20Max (With Tools) | 75.062 / 20Max (With Tools) | -1.97 |
临床工作与医疗决策
Claude Opus 5 领先 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| MedCode | 42.8637 / 64Max (With Tools) | 63.571 / 64Max (With Tools) | -20.71 |
| MedScribe | 88.818 / 66Max (With Tools) | 90.983 / 66Max (With Tools) | -2.17 |
仓库修复与多文件工程
Claude Opus 5 领先 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| NL2Repo-Bench | 586 / 16Max (With Tools) | 75.301 / 16Max (With Tools) | -17.30 |
| DeepSWE | 68.9624 / 92Max (With Tools) | 73.658 / 92Max (With Tools) | -4.69 |
客服与业务流程
胶着 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Public Benefits Bench v1.1 | 68.545 / 26Max (With Tools) | 76.931 / 26Max (With Tools) | -8.39 |
| τ³-Banking | 50.305 / 167Max (With Tools) | 48.716 / 167Max (With Tools) | +1.59 |
法律
胶着 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Legal Research Bench | 49.046 / 42Max (With Tools) | 55.291 / 42Max (With Tools) | -6.25 |
| Harvey's Legal Agent Benchmark | 8.3314 / 43Max (With Tools) | 6.6717 / 43Max (With Tools) | +1.66 |
科学知识与推理
Claude Opus 5 领先 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| MysteryMechanism | 22.9710 / 13Max (With Tools) | 37.394 / 13Max (With Tools) | -14.42 |
| CritPt | 19.1043 / 204Max (No Tools) | 29.1012 / 204Max (No Tools) | -10 |
科学计算与科研编程
胶着 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 8.1013 / 18Max (With Tools) | 305 / 18Max (With Tools) | -21.90 |
| SciCode | 599 / 134Max (No Tools) | 56.4027 / 134Max (No Tools) | +2.60 |
程序生成与编辑
Claude Opus 5 领先 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Program Bench | 1910 / 15Max (With Tools) | 376 / 15Max (With Tools) | -18 |
| Vibe Code Bench v1.1 | 78.1220 / 61Max (With Tools) | 88.406 / 61Max (With Tools) | -10.28 |
跨能力聚合指数
Claude Opus 5 领先 2/2| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Vals Index | 56.9723 / 42Max (With Tools) | 67.214 / 42Max (With Tools) | -10.24 |
| Vals Index v2.1 | 53.5115 / 36Max (task-specific tools) | 63.675 / 36Max (task-specific tools) | -10.16 |
写作与创意表达
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Creative Writing | 2,0756 / 110Normal (No Tools) | 2,1333 / 110Normal (No Tools) | -57.60 |
办公与文档流程
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| AutomationBench | 48.808 / 25Max (With Tools) | 50.306 / 25Max (With Tools) | -1.50 |
文档与图表理解
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| GDP.pdf | 11.2077 / 122Max (No Tools) | 21.6031 / 122Max (No Tools) | -10.40 |
漏洞利用
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| ExploitGym (budget unspecified) | 154 / 6Max (With Tools) | 22.102 / 6Max (With Tools) | -7.10 |
算法与竞赛编程
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| IOI (Vals v2) | 68.448 / 26Max (With Tools) | 84.336 / 26Max (With Tools) | -15.89 |
维护、优化与工程管理
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Code Migration | 44.2217 / 43Max (With Tools) | 57.473 / 43Max (With Tools) | -13.25 |
综合知识考试
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| HLE | 62.507 / 235Max (With Tools) | 63.606 / 235Max (With Tools) | -1.10 |
自主开发与终端任务
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Terminal-Bench 3.0 | 28.306 / 11Max (With Tools) | 43.301 / 11Max (With Tools) | -15 |
跨应用与工具编排
Claude Opus 5 领先 1/1| 评测项 | GLM-5.3 | Claude Opus 5 | 分差 |
|---|---|---|---|
| Agents' Last Exam | 28.5012 / 24Max (With Tools) | 28.6011 / 24Max (With Tools) | -0.10 |
规格对比
| 字段 | GLM-5.3 | Claude Opus 5 |
|---|---|---|
| 发布机构 | 智谱AI | Anthropic |
| 发布时间 | 2026-08-14 | 2026-07-24 |
| 模型类型 | 推理大模型 | 推理大模型 |
| 架构 | MoE 架构 | 稠密模型 |
| 参数规模 | 7440亿 | 暂无数据 |
| 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 | GLM-5.3 | Claude Opus 5 |
|---|---|---|
| 文本输入 | $1.4 / 1M tokens | $5 / 1M tokens |
| 文本输出 | $4.4 / 1M tokens | $25 / 1M tokens |
| 缓存读取 | $0.26 / 1M tokens | $0.5 / 1M tokens |
| 缓存写入 | 暂无公开价格 | $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)、仓库修复与多文件工程 (2/2)、科学知识与推理 (2/2)、程序生成与编辑 (2/2)、跨能力聚合指数 (2/2)、写作与创意表达 (1/1)、办公与文档流程 (1/1)、文档与图表理解 (1/1)、漏洞利用 (1/1)、算法与竞赛编程 (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)、跨应用与工具编排 (1/1)
- 胶着类目:客服与业务流程、法律、科学计算与科研编程
31 个共同 benchmark 上，Claude Opus 5 平均高出 13.01 分。
单项差距最大的 benchmark：Creative Writing — GLM-5.3 2,075，Claude Opus 5 2,133（分差 -57.60）。
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。