Gemini 3.8 FlashvsGemini 3.5 Flash
在 23 个同模式 benchmark 中，Gemini 3.8 Flash 整体领先：Gemini 3.8 Flash 领先 20 项，Gemini 3.5 Flash 领先 3 项，持平 0 项，平均分差 +7.29。另有 5 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 3.5 Flash
Google DeepMind · 2026-05-19 · 多模态大模型
Gemini 3.8 Flash20 项(87%)(13%)3 项Gemini 3.5 Flash
评测分数
按能力类目分组，每组内按分差大小排列；共 23 项。
客服与业务流程
Gemini 3.8 Flash 领先 2/3| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| SAGE | 35.0649 / 65Thinking High (With Tools) | 49.8811 / 65Thinking High (With Tools) | -14.82 |
| τ³-Banking | 44.9017 / 168Thinking High (With Tools) | 32.2054 / 168Thinking High (With Tools) | +12.70 |
| Public Benefits Bench v1.1 | 65.2913 / 26Thinking High (With Tools) | 59.4720 / 26Thinking High (With Tools) | +5.82 |
数学
Gemini 3.8 Flash 领先 2/3| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| ProofBench v1.1 (Lean 4) | 4820 / 29Thinking High (With Tools) | 3122 / 29Thinking High (With Tools) | +17 |
| FrontierMath v2 | 68.4226 / 114Thinking High (With Tools) | 62.8134 / 114Thinking High (With Tools) | +5.61 |
| FrontierMath Tier 4 v2 | 21.9547 / 71Thinking High (With Tools) | 26.8340 / 71Thinking High (With Tools) | -4.88 |
临床工作与医疗决策
胶着 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| MedScribe | 84.5023 / 68Thinking High (With Tools) | 76.5751 / 68Thinking High (With Tools) | +7.93 |
| MedCode | 48.1323 / 66Thinking High (With Tools) | 55.834 / 66Thinking High (With Tools) | -7.70 |
仓库修复与多文件工程
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| DeepSWE | 73.839 / 101Thinking High (With Tools) | 36.0690 / 101Thinking High (With Tools) | +37.77 |
| APEX-SWE | 36.3049 / 73Thinking High (With Tools) | 36.1050 / 73Thinking High (With Tools) | +0.20 |
法律
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| Legal Research Bench | 38.9423 / 46Thinking High (With Tools) | 30.7734 / 46Thinking High (With Tools) | +8.17 |
| Harvey's Legal Agent Benchmark | 1013 / 50Thinking High (With Tools) | 2.5035 / 50Thinking High (With Tools) | +7.50 |
跨能力聚合指数
Gemini 3.8 Flash 领先 2/2| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| Vals Index v2.1 | 54.8314 / 37Thinking High (With Tools) | 44.7932 / 37Thinking High (With Tools) | +10.04 |
| Vals Index | 62.2510 / 42Thinking High (With Tools) | 53.0828 / 42Thinking High (With Tools) | +9.17 |
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 3.5 Flash | 分差 |
|---|---|---|---|
| MMMU-Pro | 85.604 / 231Thinking High (No Tools) | 84.3012 / 231Thinking High (No Tools) | +1.30 |
文档与图表理解
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| GDP.pdf | 2135 / 124Thinking High (No Tools) | 19.8044 / 124Thinking High (No Tools) | +1.20 |
科学知识与推理
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| CritPt | 18.3045 / 204Thinking High (No Tools) | 13.1064 / 204Thinking High (No Tools) | +5.20 |
科学计算与科研编程
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| SciCode | 56.6025 / 135Thinking High (No Tools) | 53.9051 / 135Thinking High (No Tools) | +2.70 |
程序生成与编辑
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| Vibe Code Bench v1.1 | 78.6521 / 66Thinking High (With Tools) | 48.6841 / 66Thinking High (With Tools) | +29.97 |
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 3.5 Flash | 分差 |
|---|---|---|---|
| Code Migration | 36.5528 / 48Thinking High (With Tools) | 26.7536 / 48Thinking High (With Tools) | +9.80 |
自主开发与终端任务
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 19.7045 / 105Thinking High (With Tools) | 6.6071 / 105Thinking High (With Tools) | +13.10 |
跨能力综合测试
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| LiveBench | 75.8336 / 157Thinking High (No Tools) | 74.6446 / 157Thinking High (No Tools) | +1.19 |
金融
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | Gemini 3.5 Flash | 分差 |
|---|---|---|---|
| EMB (Excel Modeling) | 72.208 / 46Thinking High (With Tools) | 63.5524 / 46Thinking High (With Tools) | +8.65 |
测试模式不同的成绩
共 5 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 3.5 Flash |
|---|---|---|
| ARC-AGI-1 | 98.50Thinking High (No Tools) | 92.50Thinking High (With Tools) |
| SimpleBench | 82.40Thinking Medium (No Tools) | 76.70Normal (No Tools) |
| CharXiv RQ | 86.20Thinking Medium (No Tools) | 84.90Thinking (With Tools) |
| Finance Agent v2 | 61.40Thinking (With Tools) | 57.86Thinking High (With Tools) |
| HLE | 44.52Reported best (effort unspecified) | 40.20Thinking High (With Tools) |
规格对比
| 字段 | Gemini 3.8 Flash | Gemini 3.5 Flash |
|---|---|---|
| 发布机构 | Google DeepMind | Google DeepMind |
| 发布时间 | 2026-09-02 | 2026-05-19 |
| 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 3.5 Flash |
|---|---|---|
| 文本输入 | $0.75 / 1M tokens | $1.5 / 1M tokens |
| 文本输出 | $3.75 / 1M tokens | $9 / 1M tokens |
| 缓存读取 | $0.075 / 1M tokens | $0.15 / 1M tokens |
小结
- Gemini 3.8 Flash在以下类目领先:客服与业务流程 (2/3)、数学 (2/3)、仓库修复与多文件工程 (2/2)、法律 (2/2)、跨能力聚合指数 (2/2)、图像感知与视觉推理 (1/1)、文档与图表理解 (1/1)、科学知识与推理 (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)、自主开发与终端任务 (1/1)、跨能力综合测试 (1/1)、金融 (1/1)
- 胶着类目:临床工作与医疗决策
23 个同模式 benchmark 上，Gemini 3.8 Flash 平均高出 7.29 分。
单项差距最大的 benchmark：DeepSWE — Gemini 3.8 Flash 73.83，Gemini 3.5 Flash 36.06（分差 +37.77）。
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。