GPT-5.5vsGemini 3.1 Pro Preview
在 16 个共同 benchmark 中，GPT-5.5 整体领先：GPT-5.5 领先 10 项，Gemini 3.1 Pro Preview 领先 5 项，持平 1 项，平均分差 +6.98。
GPT-5.5
OpenAI · 2026-04-23 · 推理大模型
Gemini 3.1 Pro Preview
Google Deep Mind · 2026-02-20 · 多模态大模型
GPT-5.510 项(63%)持平1(31%)5 项Gemini 3.1 Pro Preview
评测分数
按能力类目分组，每组内按分差大小排列；共 16 项。
General Knowledge
GPT-5.5 领先 3/5| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| ARC-AGI-2 | 852 / 62Thinking High (No Tools) | 77.109 / 62Thinking High (No Tools) | +7.90 |
| HLE | 52.2020 / 172Thinking High (With Tools) | 51.4022 / 172Thinking High (With Tools) | +0.80 |
| LiveBench | 80.711 / 115Deep Thinking (No Tools) | 79.933 / 115Thinking High (No Tools) | +0.78 |
| GPQA Diamond | 93.606 / 187Thinking High (No Tools) | 94.303 / 187Thinking High (No Tools) | -0.70 |
| ARC-AGI-3 | 05 / 9Thinking High (No Tools) | 06 / 9Thinking High (No Tools) | 持平 |
AI Agent - Tool Usage
GPT-5.5 领先 3/4| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Terminal Bench 2.0 | 82.701 / 47Thinking High (With Tools) | 68.508 / 47Thinking High (With Tools) | +14.20 |
| TerminalBench 2.1 | 83.407 / 27Thinking High (With Tools) | 73.8016 / 27Thinking High (With Tools) | +9.60 |
| MCP-Atlas | 75.3012 / 27极高强度思考（工具） | 78.209 / 27Thinking High (With Tools) | -2.90 |
| OSWorld-Verified | 78.708 / 24Thinking High (With Tools) | 76.2011 / 24Thinking (With Tools) | +2.50 |
Coding and Software Engineer
GPT-5.5 领先 2/2| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| DeepSWE | 677 / 19极高强度思考（工具） | 1219 / 19Thinking High (With Tools) | +55 |
| SWE-Bench Pro - Public | 58.6013 / 54Thinking High (With Tools) | 54.2032 / 54Thinking High (With Tools) | +4.40 |
Math and Reasoning
GPT-5.5 领先 2/2| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| FrontierMath - Tier 4 | 35.407 / 80Thinking High (With Tools) | 16.7020 / 80Normal (No Tools) | +18.70 |
| FrontierMath | 51.702 / 60Thinking High (With Tools) | 36.9011 / 60Thinking High (No Tools) | +14.80 |
Agent Level Benchmark
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| τ²-Bench - Telecom | 985 / 35Thinking High (With Tools) | 99.301 / 35Thinking High (With Tools) | -1.30 |
AI Agent - Information Search
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| BrowseComp | 84.408 / 53Thinking High (With Tools + Internet) | 85.905 / 53Thinking High (With Tools + Internet) | -1.50 |
常识推理
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | GPT-5.5 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Simple Bench | 697 / 63Normal (No Tools) | 79.602 / 63Normal (No Tools) | -10.60 |
规格对比
| 字段 | GPT-5.5 | Gemini 3.1 Pro Preview |
|---|---|---|
| 发布机构 | OpenAI | Google Deep Mind |
| 发布时间 | 2026-04-23 | 2026-02-20 |
| 模型类型 | 推理大模型 | 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 | 1000K | 1M |
| 最大输出 | 128K | 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 | GPT-5.5 | Gemini 3.1 Pro Preview |
|---|---|---|
| 文本输入 | $0.5 / 1M tokens | $2 / 1M tokens |
| 文本输出 | $30 / 1M tokens | $12 / 1M tokens |
| 缓存读取 | $0.5 / 1M tokens | 暂无公开价格 |
| 缓存写入 | $6.25 / 1M tokens | 暂无公开价格 |
小结
- GPT-5.5在以下类目领先:General Knowledge (3/5)、AI Agent - Tool Usage (3/4)、Coding and Software Engineer (2/2)、Math and Reasoning (2/2)
- Gemini 3.1 Pro Preview在以下类目领先:Agent Level Benchmark (1/1)、AI Agent - Information Search (1/1)、常识推理 (1/1)
16 个共同 benchmark 上，GPT-5.5 平均高出 6.98 分。
单项差距最大的 benchmark：DeepSWE — GPT-5.5 67，Gemini 3.1 Pro Preview 12（分差 +55）。
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。