Claude Sonnet 4 vs Claude Sonnet 4.5
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Claude Sonnet 4 | Claude Sonnet 4.5 |
|---|---|---|
| Provider |  Anthropic | Anthropic |
| pontuação média | 44.6 | 42.1 |
| Preço de entrada | $3.00 | $3.00 |
| Preço de saída | $15.00 | $15.00 |
| Janela de contexto | 1.0M tokens (~500 books) | 1.0M tokens (~500 books) |
| Lançado em | 2025-05-22 | 2025-09-29 |
| Código aberto | Proprietary | Proprietary |
Pontuações de benchmark
16 benchmarks · Claude Sonnet 4: 0, Claude Sonnet 4.5: 16
| Benchmark | Categoria | Claude Sonnet 4 | Claude Sonnet 4.5 |
|---|---|---|---|
| ARC-AGI | reasoning | 40.0 | 63.7 |
| ARC-AGI-2 | reasoning | 5.9 | 13.6 |
| Cybench | coding | 35.0 | 60.0 |
| DeepResearch Bench | knowledge | 47.8 | 52.6 |
| FrontierMath-2025-02-28-Private | math | 4.1 | 15.2 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 0.1 | 4.2 |
| GPQA diamond | knowledge | 72.3 | 76.4 |
| GSO-Bench | coding | 4.9 | 14.7 |
| HLE | knowledge | 3.1 | 9.4 |
| MATH level 5 | math | 84.4 | 97.7 |
| OSWorld | agentic | 43.9 | 62.9 |
| OTIS Mock AIME 2024-2025 | math | 71.1 | 77.8 |
| SimpleBench | reasoning | 34.6 | 45.2 |
| SWE-Bench Verified (Bash Only) | coding | 64.9 | 70.6 |
| VPCT | knowledge | 1.0 | 9.7 |
| WeirdML | coding | 46.1 | 47.7 |