Claude Sonnet 4.5 vs Gemini 2.5 Pro
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Claude Sonnet 4.5 | Gemini 2.5 Pro |
|---|---|---|
| Provider |  Anthropic |  Google DeepMind |
| pontuação média | 42.1 | 56.2 |
| Preço de entrada | $3.00 | $1.25 |
| Preço de saída | $15.00 | $10.00 |
| Janela de contexto | 1.0M tokens (~500 books) | 1.0M tokens (~524 books) |
| Lançado em | 2025-09-29 | 2025-06-17 |
| Código aberto | Proprietary | Proprietary |
Pontuações de benchmark
17 benchmarks · Claude Sonnet 4.5: 8, Gemini 2.5 Pro: 8
| Benchmark | Categoria | Claude Sonnet 4.5 | Gemini 2.5 Pro |
|---|---|---|---|
| ARC-AGI | reasoning | 63.7 | 41.0 |
| ARC-AGI-2 | reasoning | 13.6 | 4.9 |
| Chess Puzzles | knowledge | 12.0 | 20.0 |
| DeepResearch Bench | knowledge | 52.6 | 49.7 |
| FrontierMath-2025-02-28-Private | math | 15.2 | 14.1 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 4.2 | 4.2 |
| GPQA diamond | knowledge | 76.4 | 80.4 |
| GSO-Bench | coding | 14.7 | 3.9 |
| HLE | knowledge | 9.4 | 17.7 |
| MATH level 5 | math | 97.7 | 95.6 |
| OTIS Mock AIME 2024-2025 | math | 77.8 | 84.7 |
| SimpleBench | reasoning | 45.2 | 54.9 |
| SimpleQA Verified | knowledge | 23.6 | 56.0 |
| SWE-Bench verified | coding | 71.3 | 57.6 |
| Terminal Bench | coding | 46.5 | 32.6 |
| VPCT | knowledge | 9.7 | 19.6 |
| WeirdML | coding | 47.7 | 54.0 |