Across 11 shared benchmarks, Gemini 3.1 Pro Preview leads overall: Gemini 3.1 Pro Preview wins 10, Gemini 2.5-Pro wins 1, with 0 ties and an average score difference of +30.56.
Google Deep Mind · 2026-02-20 · Multimodal model
Google Deep Mind · 2025-06-05 · Reasoning model
Grouped by capability, sorted by largest gap within each. 11 shared benchmarks.
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| ARC-AGI-2 | 77.107 / 58Thinking High (No Tools) | 4.9043 / 58thinking | +72.20 |
| HLE | 51.4012 / 149Thinking High (With Tools) | 21.6089 / 149thinking | +29.80 |
| GPQA Diamond | 94.303 / 175Thinking High (No Tools) | 86.4038 / 175thinking | +7.90 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| LiveCodeBench | 91.703 / 118Thinking High (With Tools) | 77.1032 / 118 | +14.60 |
| SWE-bench Verified | 80.607 / 103Thinking High (With Tools) | 67.2063 / 103thinking | +13.40 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| FrontierMath | 36.9011 / 60Thinking High (No Tools) | 1123 / 60 | +25.90 |
| FrontierMath - Tier 4 | 16.7020 / 80Normal (No Tools) | 2.1056 / 80Normal (No Tools) | +14.60 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 99.301 / 35Thinking High (With Tools) | 5432 / 35thinking + 使用工具 | +45.30 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| BrowseComp | 85.903 / 43Thinking High (With Tools + Internet) | 7.8042 / 43thinking + 使用工具 | +78.10 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 68.506 / 43Thinking High (With Tools) | 32.6043 / 43thinking + 使用工具 | +35.90 |
| Benchmark | Gemini 3.1 Pro Preview | Gemini 2.5-Pro | Diff |
|---|---|---|---|
| MMMU | 80.5011 / 28Thinking High (No Tools) | 829 / 28thinking | -1.50 |
| Field | Gemini 3.1 Pro Preview | Gemini 2.5-Pro |
|---|---|---|
| Publisher | Google Deep Mind | Google Deep Mind |
| Release date | 2026-02-20 | 2025-06-05 |
| Model type | Multimodal model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | 0.0 | 0.0 |
| Context length | 1M | 1000K |
| Max output | 32768 | 65536 |
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Gemini 3.1 Pro Preview | Gemini 2.5-Pro |
|---|---|---|
| Text input | $2 / 1M tokens | 1.25 美元/100 万tokens |
| Text output | $12 / 1M tokens | 10 美元/100 万tokens |
| Cache read | Not public | 0.125 美元/100 万tokens |
On average across the 11 shared benchmarks, Gemini 3.1 Pro Preview scores 30.56 higher.
Largest single-benchmark gap: BrowseComp — Gemini 3.1 Pro Preview 85.90 vs Gemini 2.5-Pro 7.80 (+78.10).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.
