Claude Sonnet 5vsGPT-5.6 Terra
在 28 个共同 benchmark 中，GPT-5.6 Terra 整体领先：Claude Sonnet 5 领先 12 项，GPT-5.6 Terra 领先 15 项，持平 1 项，平均分差 -7.24。
Claude Sonnet 5
Anthropic · 2026-06-30 · 多模态大模型
GPT-5.6 Terra
OpenAI · 2026-06-26 · 推理大模型
Claude Sonnet 512 项(43%)持平1(54%)15 项GPT-5.6 Terra
评测分数
按能力类目分组，每组内按分差大小排列；共 28 项。
客服与业务流程
Claude Sonnet 5 领先 2/3| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Public Benefits Bench v1.1 | 66.0311 / 26Max (With Tools) | 62.3816 / 26Max (With Tools) | +3.65 |
| SAGE | 48.9217 / 64Max (With Tools) | 4726 / 64Max (With Tools) | +1.92 |
| τ³-Banking | 15.70107 / 167Normal (With Tools) | 15.70107 / 167Normal (With Tools) | 持平 |
数学
GPT-5.6 Terra 领先 2/3| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| FrontierMath Tier 4 v2 | 29.2720 / 42Max (No Tools) | 70.738 / 42Max (No Tools) | -41.46 |
| FrontierMath v2 | 65.6120 / 58Max (No Tools) | 85.964 / 58Max (No Tools) | -20.35 |
| ProofBench v1.1 (Lean 4) | 779 / 28Max (With Tools) | 7410 / 28Max (With Tools) | +3 |
法律
Claude Sonnet 5 领先 3/3| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Harvey Lab-AA | 90.0714 / 44Max (With Tools) | 85.1820 / 44Max (With Tools) | +4.89 |
| Harvey's Legal Agent Benchmark | 523 / 43Max (With Tools) | 0.8333 / 43Max (With Tools) | +4.17 |
| Legal Research Bench | 41.8314 / 42Max (With Tools) | 41.3515 / 42Max (With Tools) | +0.48 |
金融
GPT-5.6 Terra 领先 2/3| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Tax Agent Bench | 62.2714 / 20Max (With Tools) | 65.2011 / 20Max (With Tools) | -2.93 |
| Finance Agent v2 | 53.9120 / 43Max (With Tools) | 54.4417 / 43Max (With Tools) | -0.53 |
| EMB (Excel Modeling) | 66.3216 / 42Max (With Tools) | 66.2017 / 42Max (With Tools) | +0.12 |
临床工作与医疗决策
胶着 2/2| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| MedScribe | 76.0553 / 66Max (With Tools) | 82.8735 / 66Max (With Tools) | -6.82 |
| MedCode | 47.5424 / 64Max (With Tools) | 43.4134 / 64Max (With Tools) | +4.13 |
仓库修复与多文件工程
Claude Sonnet 5 领先 2/2| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| SWE-Bench Pro V2 Hard | 88.204 / 11Extra-High (With Tools) | 86.306 / 11Extra-High (With Tools) | +1.90 |
| SWE-Bench Pro V2 | 93.158 / 10Extra-High (With Tools) | 92.379 / 10Extra-High (With Tools) | +0.78 |
自主开发与终端任务
GPT-5.6 Terra 领先 2/2| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 12.4255 / 98Max (With Tools) | 21.5236 / 98Max (With Tools) | -9.10 |
| CursorBench 4.0 | 34.1023 / 43Max (With Tools) | 41.3013 / 43Max (With Tools) | -7.20 |
写作与创意表达
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Creative Writing | 1,79424 / 110Normal (No Tools) | 1,85514 / 110Normal (No Tools) | -60.90 |
图像感知与视觉推理
Claude Sonnet 5 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| MMMU-Pro | 71.90118 / 229Normal (No Tools) | 66.70148 / 229Normal (No Tools) | +5.20 |
多轮记忆与持续上下文
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Context Arena | 79.5340 / 126Max (No Tools) | 92.1711 / 126Max (No Tools) | -12.64 |
机器学习工程
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| WeirdML v2 | 68.7820 / 52Thinking High (With Tools) | 78.2710 / 52Thinking High (With Tools) | -9.49 |
科学知识与推理
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| CritPt | 1.10151 / 204Normal (No Tools) | 2131 / 204Normal (No Tools) | -0.90 |
程序生成与编辑
Claude Sonnet 5 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 81.3316 / 61Max (With Tools) | 74.5923 / 61Max (With Tools) | +6.74 |
算法与竞赛编程
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| IOI (Vals v2) | 4520 / 26Max (With Tools) | 87.615 / 26Max (With Tools) | -42.61 |
维护、优化与工程管理
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Code Migration | 44.3916 / 43Max (With Tools) | 47.808 / 43Max (With Tools) | -3.41 |
综合知识考试
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| HLE-Verified | 316 / 6Reported best (effort unspecified) | 51.105 / 6Reported best (effort unspecified) | -20.10 |
跨能力聚合指数
GPT-5.6 Terra 领先 1/1| 评测项 | Claude Sonnet 5 | GPT-5.6 Terra | 分差 |
|---|---|---|---|
| Vals Index v2.1 | 51.7720 / 36Max (task-specific tools) | 53.0918 / 36Max (task-specific tools) | -1.32 |
规格对比
| 字段 | Claude Sonnet 5 | GPT-5.6 Terra |
|---|---|---|
| 发布机构 | Anthropic | OpenAI |
| 发布时间 | 2026-06-30 | 2026-06-26 |
| 模型类型 | 多模态大模型 | 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 | 1.05M |
| 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 Sonnet 5 | GPT-5.6 Terra |
|---|---|---|
| 文本输入 | $2 / 1M tokens | $2 / 1M tokens |
| 文本输出 | $10 / 1M tokens | $12 / 1M tokens |
| 缓存读取 | $0.2 / 1M tokens | $0.2 / 1M tokens |
| 缓存写入 | $2.5 / 1M tokens | $2.5 / 1M tokens |
小结
- Claude Sonnet 5在以下类目领先:客服与业务流程 (2/3)、法律 (3/3)、仓库修复与多文件工程 (2/2)、图像感知与视觉推理 (1/1)、程序生成与编辑 (1/1)
- GPT-5.6 Terra在以下类目领先:数学 (2/3)、金融 (2/3)、自主开发与终端任务 (2/2)、写作与创意表达 (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)、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)、跨能力聚合指数 (1/1)
- 胶着类目:临床工作与医疗决策
28 个共同 benchmark 上，GPT-5.6 Terra 平均高出 7.24 分。
单项差距最大的 benchmark：Creative Writing — Claude Sonnet 5 1,794，GPT-5.6 Terra 1,855（分差 -60.90）。
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。