Gemini 3.6 FlashvsClaude Sonnet 5
在 8 个同模式 benchmark 中，Claude Sonnet 5 整体领先：Gemini 3.6 Flash 领先 2 项，Claude Sonnet 5 领先 6 项，持平 0 项，平均分差 -25.55。另有 22 项测试模式不同，仅供参考。
Gemini 3.6 Flash
Google DeepMind · 2026-07-21 · 多模态大模型
Claude Sonnet 5
Anthropic · 2026-06-30 · 多模态大模型
Gemini 3.6 Flash2 项(25%)(75%)6 项Claude Sonnet 5
评测分数
按能力类目分组，每组内按分差大小排列；共 8 项。
仓库修复与多文件工程
Claude Sonnet 5 领先 2/2| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| APEX-SWE | 39.4040 / 73Thinking High (With Tools) | 43.6532 / 73Thinking High (With Tools) | -4.25 |
| DeepSWE | 46.6882 / 106Thinking High (With Tools) | 48.2381 / 106Thinking High (With Tools) | -1.55 |
写作与创意表达
Claude Sonnet 5 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| Creative Writing | 1,60441 / 112Normal (No Tools) | 1,79424 / 112Normal (No Tools) | -190.50 |
文档与图表理解
Gemini 3.6 Flash 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| GDP.pdf | 17.4053 / 126Thinking High (No Tools) | 9.4091 / 126Thinking High (No Tools) | +8 |
机器学习工程
Claude Sonnet 5 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| WeirdML v2 | 56.1034 / 53Thinking High (With Tools) | 68.7821 / 53Thinking High (With Tools) | -12.68 |
科学知识与推理
Claude Sonnet 5 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| CritPt | 10.6077 / 208Thinking High (No Tools) | 15.1061 / 208Thinking High (No Tools) | -4.50 |
科学计算与科研编程
Claude Sonnet 5 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| SciCode | 53.4055 / 137Thinking High (No Tools) | 54.3047 / 137Thinking High (No Tools) | -0.90 |
自主开发与终端任务
Gemini 3.6 Flash 领先 1/1| 评测项 | Gemini 3.6 Flash | Claude Sonnet 5 | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 7.1071 / 108Thinking High (With Tools) | 5.1077 / 108Thinking High (With Tools) | +2 |
测试模式不同的成绩
共 22 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 | Gemini 3.6 Flash | Claude Sonnet 5 |
|---|---|---|
| Terminal-Bench 2.1 | 78Thinking (With Tools) | 80.40Extra-High (With Tools) |
| IOI (Vals v2) | 35.06Thinking High (With Tools) | 45Max (With Tools) |
| Vals Index | 55.35Thinking High (With Tools) | 59.61Max (With Tools) |
| MedCode | 53.15Thinking High (With Tools) | 47.54Max (With Tools) |
| MedScribe | 79.66Thinking High (With Tools) | 76.05Max (With Tools) |
| Vibe Code Bench v1.1 | 64Thinking High (With Tools) | 81.33Max (With Tools) |
| LiveBench | 73.59Thinking High (No Tools) | 76.04Extra-High (No Tools) |
| OSWorld-Verified | 83Thinking (With Tools) | 81.20Extra-High (With Tools) |
| EMB (Excel Modeling) | 65.41Thinking High (With Tools) | 66.32Max (With Tools) |
| Finance Agent v2 | 56.30Thinking High (With Tools) | 53.91Max (With Tools) |
| Harvey Lab-AA | 85.15Thinking High (With Tools) | 90.07Max (With Tools) |
| Harvey's Legal Agent Benchmark | 3.33Thinking High (With Tools) | 5Max (With Tools) |
| Legal Research Bench | 25Thinking High (With Tools) | 41.83Max (With Tools) |
| Code Migration | 30.93Thinking High (With Tools) | 44.39Max (With Tools) |
| FrontierMath Tier 4 v2 | 21.95Thinking High (With Tools) | 29.27Max (With Tools) |
| FrontierMath v2 | 58.95Thinking High (With Tools) | 65.61Max (With Tools) |
| Context Arena | 88.78Thinking High (No Tools) | 79.53Max (No Tools) |
| GPQA Diamond | 94.13Thinking High (No Tools) | 90.53Extra-High (No Tools) |
| Public Benefits Bench v1.1 | 56.56Thinking High (With Tools) | 66.03Max (With Tools) |
| SAGE | 48.69Thinking High (With Tools) | 48.92Max (With Tools) |
| τ³-Banking | 29.90Thinking High (With Tools) | 15.70Normal (With Tools) |
| MMMU-Pro | 83.20Thinking High (No Tools) | 71.90Normal (No Tools) |
规格对比
| 字段 | Gemini 3.6 Flash | Claude Sonnet 5 |
|---|---|---|
| 发布机构 | Google DeepMind | Anthropic |
| 发布时间 | 2026-07-21 | 2026-06-30 |
| 模型类型 | 多模态大模型 | 多模态大模型 |
| 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 | 1M |
| 最大输出 | 64K | 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 | Gemini 3.6 Flash | Claude Sonnet 5 |
|---|---|---|
| 文本输入 | $1.5 / 1M tokens | $2 / 1M tokens |
| 文本输出 | $7.5 / 1M tokens | $10 / 1M tokens |
| 缓存读取 | $0.15 / 1M tokens | $0.2 / 1M tokens |
| 缓存写入 | 暂无公开价格 | $2.5 / 1M tokens |
小结
- Gemini 3.6 Flash在以下类目领先:文档与图表理解 (1/1)、自主开发与终端任务 (1/1)
- Claude Sonnet 5在以下类目领先:仓库修复与多文件工程 (2/2)、写作与创意表达 (1/1)、机器学习工程 (1/1)、科学知识与推理 (1/1)、科学计算与科研编程 (1/1)
8 个同模式 benchmark 上，Claude Sonnet 5 平均高出 25.55 分。
单项差距最大的 benchmark：Creative Writing — Gemini 3.6 Flash 1,604，Claude Sonnet 5 1,794（分差 -190.50）。
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。