Gemini 3.8 FlashvsGemini 3.7 Flash
在 35 个同模式 benchmark 中，Gemini 3.8 Flash 整体领先：Gemini 3.8 Flash 领先 24 项，Gemini 3.7 Flash 领先 11 项，持平 0 项，平均分差 +1.14。另有 2 项测试模式不同，仅供参考。
Gemini 3.8 Flash
Google DeepMind · 2026-09-02 · 多模态大模型
Gemini 3.7 Flash
Google DeepMind · 2026-08-13 · 多模态大模型
Gemini 3.8 Flash24 项(69%)(31%)11 项Gemini 3.7 Flash
评测分数
按能力类目分组，每组内按分差大小排列；共 35 项。
仓库修复与多文件工程
Gemini 3.7 Flash 领先 2/3| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| DeepSWE | 73.839 / 101Thinking High (With Tools) | 65.2744 / 101Thinking High (With Tools) | +8.56 |
| APEX-SWE | 36.3049 / 73Thinking High (With Tools) | 4233 / 73Thinking High (With Tools) | -5.70 |
| FrontierSWE v2 | 19.6316 / 22Max (With Tools) | 20.2615 / 22Max (With Tools) | -0.63 |
数学
Gemini 3.7 Flash 领先 3/3| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| FrontierMath Tier 4 v2 | 21.9547 / 71Thinking High (With Tools) | 36.5931 / 71Thinking High (With Tools) | -14.63 |
| ProofBench v1.1 (Lean 4) | 4820 / 29Thinking High (With Tools) | 5813 / 29Thinking High (With Tools) | -10 |
| FrontierMath v2 | 68.4226 / 114Thinking High (With Tools) | 71.5823 / 114Thinking High (With Tools) | -3.16 |
生物与基因
Gemini 3.8 Flash 领先 3/3| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| BioMysteryBench (Human-difficult) | 56.501 / 2Thinking Medium (With Tools + Internet) | 43.502 / 2Thinking Medium (With Tools + Internet) | +13 |
| LABBench2 | 86.202 / 3Thinking Medium (With Tools + Internet) | 82.103 / 3Thinking Medium (With Tools + Internet) | +4.10 |
| BioMysteryBench (Human-solvable) | 88.801 / 2Thinking Medium (With Tools + Internet) | 87.102 / 2Thinking Medium (With Tools + Internet) | +1.70 |
临床工作与医疗决策
胶着 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| MedCode | 48.1323 / 66Thinking High (With Tools) | 53.397 / 66Thinking High (With Tools) | -5.26 |
| MedScribe | 84.5023 / 68Thinking High (With Tools) | 83.9427 / 68Thinking High (With Tools) | +0.56 |
客服与业务流程
胶着 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| SAGE | 35.0649 / 65Thinking High (With Tools) | 49.2316 / 65Thinking High (With Tools) | -14.17 |
| τ³-Banking | 45.8014 / 168Thinking Medium (With Tools) | 35.5046 / 168Thinking Medium (With Tools) | +10.30 |
文档与图表理解
胶着 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| GDP.pdf | 22.8028 / 124Thinking Medium (No Tools) | 341 / 124Thinking Medium (No Tools) | -11.20 |
| CharXiv RQ | 86.208 / 23Thinking Medium (No Tools) | 84.5014 / 23Thinking Medium (No Tools) | +1.70 |
法律
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Legal Research Bench | 38.9423 / 46Thinking High (With Tools) | 34.6231 / 46Thinking High (With Tools) | +4.32 |
| Harvey's Legal Agent Benchmark | 1013 / 50Thinking High (With Tools) | 8.7515 / 50Thinking High (With Tools) | +1.25 |
科学计算与科研编程
胶着 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 12.4010 / 18High Thinking (With Tools) | 5.7016 / 18High Thinking (With Tools) | +6.70 |
| SciCode | 55.1039 / 135Thinking Medium (No Tools) | 59.805 / 135Thinking Medium (No Tools) | -4.70 |
自主开发与终端任务
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 19.7045 / 105Thinking High (With Tools) | 13.6056 / 105Thinking High (With Tools) | +6.10 |
| Terminal-Bench 2.1 | 89.404 / 63Thinking (With Tools) | 85.8017 / 63Thinking (With Tools) | +3.60 |
跨能力聚合指数
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Vals Index v2.1 | 54.8314 / 37Thinking High (With Tools) | 51.2723 / 37Thinking High (With Tools) | +3.56 |
| Vals Index | 62.2510 / 42Thinking High (With Tools) | 59.3116 / 42Thinking High (With Tools) | +2.94 |
金融
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Tax Agent Bench | 66.779 / 21Thinking High (With Tools) | 57.6620 / 21Thinking High (With Tools) | +9.11 |
| EMB (Excel Modeling) | 72.208 / 46Thinking High (With Tools) | 71.3311 / 46Thinking High (With Tools) | +0.87 |
写作与创意表达
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Creative Writing | 1,74826 / 112Normal (No Tools) | 1,72328 / 112Normal (No Tools) | +24.70 |
图像感知与视觉推理
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| MMMU-Pro | 85.604 / 231Thinking High (No Tools) | 85.505 / 231Thinking High (No Tools) | +0.10 |
抽象归纳与泛化
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| ARC-AGI-1 | 98.501 / 226Thinking High (No Tools) | 95.5032 / 226Thinking High (No Tools) | +3 |
视频理解
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| LVBench | 87.103 / 5Thinking Medium (No Tools) | 85.404 / 5Thinking Medium (No Tools) | +1.70 |
科学知识与推理
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| CritPt | 18.3045 / 204Thinking High (No Tools) | 14.3061 / 204Thinking High (No Tools) | +4 |
程序生成与编辑
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 78.6521 / 66Thinking High (With Tools) | 70.3930 / 66Thinking High (With Tools) | +8.26 |
算法与竞赛编程
Gemini 3.7 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| IOI (Vals v2) | 56.9415 / 30Thinking High (With Tools) | 67.8312 / 30Thinking High (With Tools) | -10.89 |
维护、优化与工程管理
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| Code Migration | 36.5528 / 48Thinking High (With Tools) | 34.8030 / 48Thinking High (With Tools) | +1.75 |
综合知识考试
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| HLE-Verified | 54.901 / 6Thinking Medium (No Tools) | 53.604 / 6Thinking Medium (No Tools) | +1.30 |
跨能力综合测试
Gemini 3.7 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.7 Flash | 分差 |
|---|---|---|---|
| LiveBench | 75.8336 / 157Thinking High (No Tools) | 78.8318 / 157Thinking High (No Tools) | -3 |
测试模式不同的成绩
共 2 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 3.8 Flash | Gemini 3.7 Flash |
|---|---|---|
| OSWorld 2.0 | 59Thinking (With Tools) | 47.90Thinking Medium (With Tools) |
| Finance Agent v2 | 61.40Thinking (With Tools) | 59.04Thinking High (With Tools) |
规格对比
| 字段 | Gemini 3.8 Flash | Gemini 3.7 Flash |
|---|---|---|
| 发布机构 | Google DeepMind | Google DeepMind |
| 发布时间 | 2026-09-02 | 2026-08-13 |
| 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 | 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 | Gemini 3.8 Flash | Gemini 3.7 Flash |
|---|---|---|
| 文本输入 | $0.75 / 1M tokens | $0.75 / 1M tokens |
| 文本输出 | $3.75 / 1M tokens | $3.75 / 1M tokens |
| 缓存读取 | $0.075 / 1M tokens | $0.075 / 1M tokens |
小结
- Gemini 3.8 Flash在以下类目领先:生物与基因 (3/3)、法律 (2/2)、自主开发与终端任务 (2/2)、跨能力聚合指数 (2/2)、金融 (2/2)、写作与创意表达 (1/1)、图像感知与视觉推理 (1/1)、抽象归纳与泛化 (1/1)、视频理解 (1/1)、科学知识与推理 (1/1)、程序生成与编辑 (1/1)、维护、优化与工程管理 (1/1)、综合知识考试 (1/1)
- Gemini 3.7 Flash在以下类目领先:仓库修复与多文件工程 (2/3)、数学 (3/3)、算法与竞赛编程 (1/1)、跨能力综合测试 (1/1)
- 胶着类目:临床工作与医疗决策、客服与业务流程、文档与图表理解、科学计算与科研编程
35 个同模式 benchmark 上，Gemini 3.8 Flash 平均高出 1.14 分。
单项差距最大的 benchmark：Creative Writing — Gemini 3.8 Flash 1,748，Gemini 3.7 Flash 1,723（分差 +24.70）。
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。