Opus 4.7vsGemini 3.1 Pro Preview
在 11 个同模式 benchmark 中，Opus 4.7 整体领先：Opus 4.7 领先 7 项，Gemini 3.1 Pro Preview 领先 4 项，持平 0 项，平均分差 +47.11。另有 27 项测试模式不同，仅供参考。
Opus 4.7
Anthropic · 2026-04-16 · 推理大模型
Gemini 3.1 Pro Preview
Google DeepMind · 2026-02-20 · 多模态大模型
Opus 4.77 项(64%)(36%)4 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 11 项。
仓库修复与多文件工程
Opus 4.7 领先 2/2| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| FrontierSWE v1 | 56.258 / 17Historical report (mode unspecified) | 34.3811 / 17Historical report (mode unspecified) | +21.88 |
| APEX-SWE | 41.3436 / 73Thinking High (With Tools) | 33.9055 / 73Thinking High (With Tools) | +7.44 |
写作与创意表达
Opus 4.7 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Creative Writing | 1,91412 / 112Normal (No Tools) | 1,49157 / 112Normal (No Tools) | +423 |
多轮记忆与持续上下文
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Context Arena | 44.9096 / 126Thinking High (No Tools) | 84.7325 / 126Thinking High (No Tools) | -39.83 |
常识推理
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| SimpleBench | 61.7030 / 95Normal (No Tools) | 79.609 / 95Normal (No Tools) | -17.90 |
机器学习工程
Opus 4.7 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| WeirdML v2 | 76.4013 / 53Normal (With Tools) | 72.1017 / 53Normal (With Tools) | +4.30 |
维护、优化与工程管理
Opus 4.7 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| GSO | 44.121 / 22Normal (With Tools) | 22.5510 / 22Normal (With Tools) | +21.57 |
综合偏好评测
Opus 4.7 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Text Arena (Coding) | 1,5627 / 36Normal (No Tools) | 1,46124 / 36Normal (No Tools) | +100.90 |
自主开发与终端任务
Opus 4.7 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| Terminal-Bench 2.1 | 69.7045 / 62Thinking High (With Tools) | 65.8449 / 62Thinking High (With Tools) | +3.86 |
跨应用与工具编排
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| PinchBench v2 | 76.0115 / 45Reported best (effort unspecified) | 81.019 / 45Reported best (effort unspecified) | -5 |
跨能力综合测试
Gemini 3.1 Pro Preview 领先 1/1| 评测项 | Opus 4.7 | Gemini 3.1 Pro Preview | 分差 |
|---|---|---|---|
| LiveBench | 74.8945 / 157Thinking High (No Tools) | 76.9529 / 157Thinking High (No Tools) | -2.06 |
测试模式不同的成绩
共 27 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 | Opus 4.7 | Gemini 3.1 Pro Preview |
|---|---|---|
| Terminal Bench 2.0 | 69.40Extended (With Tools) | 68.50Thinking High (With Tools) |
| Terminal Bench Hard | 54.50Normal (With Tools) | 53.80Thinking (With Tools) |
| Vals Index | 56.11Max (With Tools) | 41.90Thinking High (With Tools) |
| MedCode | 54.86Max (With Tools) | 59.06Thinking High (With Tools) |
| MedScribe | 82.95Max (With Tools) | 76.11Thinking High (With Tools) |
| Vibe Code Bench v1.1 | 71Max (With Tools) | 32.03Thinking High (With Tools) |
| AA-AnalystAgent | 43.75Max (With Tools + Internet) | 41.25Thinking (With Tools + Internet) |
| OSWorld-Verified | 78Extended (With Tools) | 76.20Thinking (With Tools) |
| BrowseComp | 79.30Extended (With Tools) | 85.90Thinking High (With Tools + Internet) |
| EMB (Excel Modeling) | 63.73Max (With Tools) | 52.62Thinking High (With Tools) |
| Finance Agent v2 | 51.51Max (With Tools) | 42.98Thinking High (With Tools) |
| HLE | 54.70Extended (With Tools) | 51.40Thinking High (With Tools) |
| Legal Research Bench | 38.46Max (With Tools) | 20.67Thinking High (With Tools) |
| Code Migration | 43.88Max (With Tools) | 17.31Thinking High (With Tools) |
| FrontierMath | 43.80Extra-High (No Tools) | 36.90Thinking High (No Tools) |
| FrontierMath - Tier 4 | 22.90Extra-High (No Tools) | 16.70Normal (No Tools) |
| FrontierMath Tier 4 v2 | 31.71Max (With Tools) | 26.83Reported best (effort unspecified) |
| FrontierMath v2 | 70.18Max (With Tools) | 59.65Reported best (effort unspecified) |
| EBR-bench | 19.05Max (With Tools) | 14.29Thinking (With Tools) |
| SWE-Bench Pro - Public | 64.30Extended (With Tools) | 54.20Thinking High (With Tools) |
| SWE-bench Verified | 87.60Extended (With Tools) | 80.60Thinking High (With Tools) |
| CritPt | 5.10Normal (No Tools) | 17.70Thinking (No Tools) |
| GPQA Diamond | 94.20Extended (No Tools) | 94.30Thinking High (No Tools) |
| SAGE | 56.10Max (With Tools) | 48.68Thinking High (With Tools) |
| τ³-Banking | 40.21Max (With Tools) | 26.03Thinking High (With Tools) |
| MCP-Atlas | 79.10Deep Thinking (With Tools) | 78.20Thinking High (With Tools) |
| MMMU-Pro | 76.40Normal (No Tools) | 82.40Thinking (No Tools) |
规格对比
| 字段 | Opus 4.7 | Gemini 3.1 Pro Preview |
|---|---|---|
| 发布机构 | Anthropic | Google DeepMind |
| 发布时间 | 2026-04-16 | 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 | Opus 4.7 | Gemini 3.1 Pro Preview |
|---|---|---|
| 文本输入 | $5 / 1M tokens | $2 / 1M tokens |
| 文本输出 | $25 / 1M tokens | $12 / 1M tokens |
| 缓存读取 | $0.5 / 1M tokens | $0.2 / 1M tokens |
| 缓存写入 | $6.25 / 1M tokens | 暂无公开价格 |
小结
- Opus 4.7在以下类目领先:仓库修复与多文件工程 (2/2)、写作与创意表达 (1/1)、机器学习工程 (1/1)、维护、优化与工程管理 (1/1)、综合偏好评测 (1/1)、自主开发与终端任务 (1/1)
- Gemini 3.1 Pro Preview在以下类目领先:多轮记忆与持续上下文 (1/1)、常识推理 (1/1)、跨应用与工具编排 (1/1)、跨能力综合测试 (1/1)
11 个同模式 benchmark 上，Opus 4.7 平均高出 47.11 分。
单项差距最大的 benchmark：Creative Writing — Opus 4.7 1,914，Gemini 3.1 Pro Preview 1,491（分差 +423）。
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。