Gemini 3 Pro vs Gemini 2.5 Pro
Côte à côte. Chaque métrique. Chaque benchmark.
| Type | Gemini 3 Pro | Gemini 2.5 Pro |
|---|---|---|
| Provider |  Google DeepMind | Google DeepMind |
| score moyen | 60.5 | 56.2 |
| Prix d'entrée | - | $1.25 |
| Prix de sortie | - | $10.00 |
| Fenêtre de contexte | - | 1.0M tokens (~524 books) |
| Sorti le | 2024-01-01 | 2025-06-17 |
| Code source ouvert | Proprietary | Proprietary |
Scores de benchmark
26 benchmarks · Gemini 3 Pro: 26, Gemini 2.5 Pro: 0
| Benchmark | Catégorie | Gemini 3 Pro | Gemini 2.5 Pro |
|---|---|---|---|
| ARC-AGI | reasoning | 75.0 | 41.0 |
| ARC-AGI-2 | reasoning | 31.1 | 4.9 |
| Artificial Analysis — Agentic Index | speed | 45.0 | 32.7 |
| Artificial Analysis — Coding Index | speed | 39.4 | 31.9 |
| Artificial Analysis — Quality Index | speed | 41.3 | 34.6 |
| Chatbot Arena Elo — Coding | arena | 1437.6 | 1202.0 |
| Chatbot Arena Elo — Overall | arena | 1486.2 | 1448.2 |
| Chess Puzzles | knowledge | 31.0 | 20.0 |
| FrontierMath-2025-02-28-Private | math | 37.6 | 14.1 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 18.8 | 4.2 |
| GeoBench | knowledge | 84.0 | 81.0 |
| GPQA diamond | knowledge | 90.2 | 80.4 |
| GSO-Bench | coding | 18.6 | 3.9 |
| HELM — GPQA | knowledge | 80.3 | 74.9 |
| HELM — IFEval | language | 87.6 | 84.0 |
| HELM — MMLU-Pro | knowledge | 90.3 | 86.3 |
| HELM — Omni-MATH | math | 55.6 | 41.6 |
| HELM — WildBench | reasoning | 85.9 | 85.7 |
| HLE | knowledge | 34.4 | 17.7 |
| OTIS Mock AIME 2024-2025 | math | 91.4 | 84.7 |
| SimpleBench | reasoning | 71.7 | 54.9 |
| SimpleQA Verified | knowledge | 72.9 | 56.0 |
| SWE-Bench verified | coding | 72.9 | 57.6 |
| Terminal Bench | coding | 69.4 | 32.6 |
| VPCT | knowledge | 86.5 | 19.6 |
| WeirdML | coding | 69.9 | 54.0 |