Head-to-head正面对比
Claude Opus 5 vs Gemini 3.1 Pro Preview
Anthropic meets Google: sourced benchmark scores side by side, dimension by dimension, with prices and specs. Every conclusion below is derived from the published dataset — no opinions, no sponsorship.
Anthropic 对 Google:带来源的 benchmark 成绩逐项并排,附价格与规格。下面每条结论都从公开数据集推导——没有主观意见,没有厂商赞助。
◈
The verdict, derived from data结论(全部由数据推导)
- On overall score they are nearly tied (83.3 vs 86.1) — decide on price, context and workload mix.总分几乎打平(83.3 对 86.1)——按价格、上下文和任务结构来选。
- Claude Opus 5 is stronger in reasoning, agent.Claude Opus 5 在推理、智能体上更强。
- Gemini 3.1 Pro Preview is stronger in knowledge, math.Gemini 3.1 Pro Preview 在知识、数学上更强。
- Gemini 3.1 Pro Preview is 2.1× cheaper on blended price ($7.00 vs $15 per 1M tokens, in/out average).混合价(输入输出均值)上 Gemini 3.1 Pro Preview 便宜 2.1 倍:$7.00 对 $15/百万 token。
● Claude Opus 5 ● Gemini 3.1 Pro Preview · missing dimensions count as 0 in the chart缺数据的维度在图上按 0 计
≡
Spec by spec逐项规格
| Claude Opus 5 | Gemini 3.1 Pro Preview | |
|---|---|---|
| Provider厂商 | Anthropic | |
| Released发布 | 2026-07-24 | 2026-02-19 |
| Intelligence Score智能评分 | 83.3 | 86.1 |
| Benchmark coveragebenchmark 覆盖 | 100% | 100% |
| Reasoning推理 | 96.4 | 90.5 |
| Coding代码 | 85.9 | 84.6 |
| Knowledge知识 | 88.5 | 100 |
| Math数学 | 67.2 | 96.3 |
| Agent智能体 | 62 | 50.6 |
| Preference偏好 | 66.5 | 65.8 |
| API input $/1M输入价 $/百万 | $5.00 | $2.00 |
| API output $/1M输出价 $/百万 | $25 | $12 |
| Blended $/1M混合价 $/百万 | $15 | $7.00 |
| Value (score per $)性价比(分数/美元) | 5.6 | 12.3 |
| Context window上下文 | 1M | 1M |
| Max output最大输出 | 128K | 66K |
| Reasoning tiers推理档位 | low / medium / high / x-high | minimal / low / medium / high |
| Open weights开放权重 | No否 | No否 |
| CN-direct国内直连 | No否 | No否 |
| Free tier免费层 | No否 | No否 |
⚔
Benchmark by benchmark逐项 benchmark
| Benchmark基准 | Claude Opus 5 | Gemini 3.1 Pro Preview |
|---|---|---|
| Humanity's Last Exam | 54.4⚡max | 46.44⚡high |
| GPQA Diamond | 93.7 | 95.5 |
| MMLU-Pro | 81.97 | 92.6 |
| SWE-bench Verified | 96 | 80.6 |
| AIME 2025 | 97 | 93.3 |
| LiveCodeBench | 89.03 | 88.48 |
| FrontierMath | 31.25 | 93.7 |
| Terminal-Bench | 43.5⚡max | 68.5 |
| τ²-bench | 48.71⚡max | 26.03⚡high |
| LMArena (Chatbot Arena) | 1489⚡max | 1486 |
| WebArena | 68.7 | 69 |
Claude Opus 5 dossier → 档案 → · Gemini 3.1 Pro Preview dossier → 档案 → · How we score评分方法