Gemini 3.8 FlashvsGLM-5.3-Flash
在 3 个同模式 benchmark 中，Gemini 3.8 Flash 整体领先：Gemini 3.8 Flash 领先 2 项，GLM-5.3-Flash 领先 1 项，持平 0 项，平均分差 +4.82。另有 22 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
GLM-5.3-Flash
智谱AI · 2026-08-26 · 多模态大模型
Gemini 3.8 Flash2 项(67%)(33%)1 项GLM-5.3-Flash
评测分数
按能力类目分组，每组内按分差大小排列；共 3 项。
仓库修复与多文件工程
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | GLM-5.3-Flash | 分差 |
|---|---|---|---|
| FrontierSWE v2 | 19.6316 / 22Max (With Tools) | 18.1317 / 22Max (With Tools) | +1.50 |
抽象归纳与泛化
Gemini 3.8 Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | GLM-5.3-Flash | 分差 |
|---|---|---|---|
| ARC-AGI-1 | 98.501 / 226Thinking High (No Tools) | 71.83132 / 226Thinking High (No Tools) | +26.67 |
自主开发与终端任务
GLM-5.3-Flash 领先 1/1| 评测项 | Gemini 3.8 Flash | GLM-5.3-Flash | 分差 |
|---|---|---|---|
| Terminal-Bench 4.0 | 19.0947 / 105Thinking (With Tools) | 32.8033 / 105Thinking (With Tools) | -13.71 |
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 3.8 Flash | GLM-5.3-Flash |
|---|---|---|
| Terminal-Bench 2.1 | 89.40Thinking (With Tools) | 84.30Max (With Tools) |
| IOI (Vals v2) | 56.94Thinking High (With Tools) | 52.50Max (With Tools) |
| Vals Index | 62.25Thinking High (With Tools) | 47.22Max (With Tools) |
| MedScribe | 84.50Thinking High (With Tools) | 88.94Max (With Tools) |
| Vibe Code Bench v1.1 | 78.65Thinking High (With Tools) | 30.76Max (With Tools) |
| LiveBench | 75.83Thinking High (No Tools) | 71.59Reported best (effort unspecified) |
| CharXiv RQ | 86.20Thinking Medium (No Tools) | 89.40Max (With Tools) |
| GDP.pdf | 22.80Thinking Medium (No Tools) | 15.40Thinking (No Tools) |
| EMB (Excel Modeling) | 72.20Thinking High (With Tools) | 55.93Max (With Tools) |
| Finance Agent v2 | 61.40Thinking (With Tools) | 57.85Max (With Tools) |
| HLE | 44.52Reported best (effort unspecified) | 55.30Max (With Tools) |
| Harvey's Legal Agent Benchmark | 10Thinking High (With Tools) | 6.67Max (With Tools) |
| Legal Research Bench | 38.94Thinking High (With Tools) | 45.19Max (With Tools) |
| Code Migration | 36.55Thinking High (With Tools) | 20.52Max (With Tools) |
| FrontierMath Tier 4 v2 | 21.95Thinking High (With Tools) | 17.07Max (With Tools) |
| FrontierMath v2 | 68.42Thinking High (With Tools) | 55.79Max (With Tools) |
| ProofBench v1.1 (Lean 4) | 48Thinking High (With Tools) | 21Max (With Tools) |
| APEX-SWE | 36.30Thinking High (With Tools) | 46.10Max (With Tools) |
| DeepSWE | 73.83Thinking High (With Tools) | 63.39Max (With Tools) |
| SciCode | 56.60Thinking High (No Tools) | 51.60Thinking (No Tools) |
| CritPt | 18.30Thinking High (No Tools) | 15.40Thinking (No Tools) |
| τ³-Banking | 45.80Thinking Medium (With Tools) | 47.20Thinking (With Tools) |
规格对比
| 字段 | Gemini 3.8 Flash | GLM-5.3-Flash |
|---|---|---|
| 发布机构 | Google DeepMind | 智谱AI |
| 发布时间 | 2026-09-02 | 2026-08-26 |
| 模型类型 | 多模态大模型 | 多模态大模型 |
| 架构 | 稠密模型 | MoE 架构 |
| 参数规模 | 暂无数据 | 3200亿 |
| 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 3.8 Flash | GLM-5.3-Flash |
|---|---|---|
| 文本输入 | $0.75 / 1M tokens | $0.15 / 1M tokens |
| 文本输出 | $3.75 / 1M tokens | $0.5 / 1M tokens |
| 缓存读取 | $0.075 / 1M tokens | $0.03 / 1M tokens |
小结
- Gemini 3.8 Flash在以下类目领先:仓库修复与多文件工程 (1/1)、抽象归纳与泛化 (1/1)
- GLM-5.3-Flash在以下类目领先:自主开发与终端任务 (1/1)
3 个同模式 benchmark 上，Gemini 3.8 Flash 平均高出 4.82 分。
单项差距最大的 benchmark：ARC-AGI-1 — Gemini 3.8 Flash 98.50，GLM-5.3-Flash 71.83（分差 +26.67）。
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。