Gemini 3.1 Pro PreviewvsClaude Opus 4.6
在 23 个共同 benchmark 中，Gemini 3.1 Pro Preview 整体领先：Gemini 3.1 Pro Preview 领先 13 项，Claude Opus 4.6 领先 9 项，持平 1 项，平均分差 -2.92。
Gemini 3.1 Pro Preview
Google Deep Mind · 2026-02-20 · 多模态大模型
Claude Opus 4.6
Anthropic · 2026-02-05 · 推理大模型
Gemini 3.1 Pro Preview13 项(57%)持平1(39%)9 项Claude Opus 4.6
评测分数
按能力类目分组，每组内按分差大小排列；共 23 项。
Coding and Software Engineer
Claude Opus 4.6 领先 4/6| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| Text Arena (Coding) | 1,46123 / 35Normal (No Tools) | 1,5558 / 35Normal (No Tools) | -93.86 |
| LiveCodeBench | 91.703 / 126Thinking High (With Tools) | 7640 / 126Extended (no tools) | +15.70 |
| SWE-Bench Pro - Commercial | 32.203 / 3Thinking (With Tools) | 47.101 / 3Thinking (With Tools) | -14.90 |
| GSO | 22.5510 / 21Normal (With Tools) | 33.335 / 21Normal (With Tools) | -10.78 |
| WeirdML v2 | 72.1017 / 52Normal (With Tools) | 65.9023 / 52Normal (With Tools) | +6.20 |
| SWE-bench Verified | 80.6011 / 114Thinking High (With Tools) | 80.8410 / 114Extended (with tools) | -0.24 |
General Knowledge
Gemini 3.1 Pro Preview 领先 3/5| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| ARC-AGI-2 | 77.109 / 62Thinking High (No Tools) | 66.3017 / 62Extended (no tools) | +10.80 |
| LiveBench | 79.933 / 115Thinking High (No Tools) | 76.338 / 115Thinking High (No Tools) | +3.60 |
| HLE | 51.4024 / 181Thinking High (With Tools) | 5320 / 181Extended (with tools, internet) | -1.60 |
| MMLU | 92.603 / 66Thinking High (No Tools) | 91.057 / 66Extended (no tools) | +1.55 |
| ARC-AGI-3 | 06 / 9Thinking High (No Tools) | 04 / 9最高（无工具） | 持平 |
AI Agent - Tool Usage
Gemini 3.1 Pro Preview 领先 3/3| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| OSWorld-Verified | 76.2012 / 26Thinking (With Tools) | 72.7016 / 26Extended (with tools) | +3.50 |
| Terminal Bench 2.0 | 68.508 / 48Thinking High (With Tools) | 65.4011 / 48Extended (with tools) | +3.10 |
| MCP-Atlas | 78.2012 / 38Thinking High (With Tools) | 76.8013 / 38Deep Thinking (With Tools) | +1.40 |
Agent Level Benchmark
胶着 2/2| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| τ²-Bench | 90.802 / 43Thinking High (With Tools) | 91.891 / 43Extended (with tools) | -1.09 |
| τ²-Bench - Telecom | 99.301 / 35Thinking High (With Tools) | 99.252 / 35Extended (with tools) | +0.05 |
Math and Reasoning
Claude Opus 4.6 领先 2/2| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| FrontierMath - Tier 4 | 16.7020 / 80Normal (No Tools) | 22.9012 / 80最高（无工具） | -6.20 |
| FrontierMath | 36.9011 / 60Thinking High (No Tools) | 40.707 / 60最高（无工具） | -3.80 |
AI Agent - Information Search
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| BrowseComp | 85.905 / 54Thinking High (With Tools + Internet) | 8411 / 54Thinking (With Tools + Internet) | +1.90 |
Claw-style Agent Evaluation
Claude Opus 4.6 领先 1/1| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| Pinch Bench | 86.7011 / 38Thinking (With Tools) | 87.408 / 38Thinking (With Tools) | -0.70 |
Multimodal Understanding
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| MMMU | 80.5012 / 29Thinking High (No Tools) | 77.3016 / 29Extended (with tools) | +3.20 |
常识推理
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| SimpleBench | 79.602 / 67Normal (No Tools) | 67.609 / 67Normal (No Tools) | +12 |
科学与综合推理
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Gemini 3.1 Pro Preview | Claude Opus 4.6 | 分差 |
|---|---|---|---|
| GPQA Diamond | 94.304 / 226Thinking High (No Tools) | 91.3128 / 226Extended (no tools) | +2.99 |
规格对比
| 字段 | Gemini 3.1 Pro Preview | Claude Opus 4.6 |
|---|---|---|
| 发布机构 | Google Deep Mind | Anthropic |
| 发布时间 | 2026-02-20 | 2026-02-05 |
| 模型类型 | 多模态大模型 | 推理大模型 |
| 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 | 1000K |
| 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 3.1 Pro Preview | Claude Opus 4.6 |
|---|---|---|
| 文本输入 | $2 / 1M tokens | $0.5 / 1M tokens |
| 文本输出 | $12 / 1M tokens | $25 / 1M tokens |
| 缓存读取 | 暂无公开价格 | $0.5 / 1M tokens |
| 缓存写入 | 暂无公开价格 | $10 / 1M tokens |
小结
- Gemini 3.1 Pro Preview在以下类目领先:General Knowledge (3/5)、AI Agent - Tool Usage (3/3)、AI Agent - Information Search (1/1)、Multimodal Understanding (1/1)、常识推理 (1/1)、科学与综合推理 (1/1)
- Claude Opus 4.6在以下类目领先:Coding and Software Engineer (4/6)、Math and Reasoning (2/2)、Claw-style Agent Evaluation (1/1)
- 胶着类目:Agent Level Benchmark
23 个共同 benchmark 上，Claude Opus 4.6 平均高出 2.92 分。
单项差距最大的 benchmark：Text Arena (Coding) — Gemini 3.1 Pro Preview 1,461，Claude Opus 4.6 1,555（分差 -93.86）。
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。