Claude Sonnet 4
por Anthropic · Lançado em 2025-05-22
36.9
pontuação média
$3.00/1M
Preço de entrada
$15.00/1M
Preço de saída
200K tokens (~100 books)
Janela de contexto
multimodal
Tipo
Tested on 20 benchmarks with 36.9% average. Top scores: MATH level 5 (84.4%), GPQA diamond (72.3%), OTIS Mock AIME 2024-2025 (71.1%).
Pontuações de benchmark
| Benchmark | Categoria | Pontuação | Bar |
|---|---|---|---|
| MATH level 5 | math | 84.4 | |
| GPQA diamond | knowledge | 72.3 | |
| OTIS Mock AIME 2024-2025 | math | 71.1 | |
| SWE-Bench Verified (Bash Only) | coding | 64.9 | |
| Aider polyglot | coding | 61.3 | |
| DeepResearch Bench | knowledge | 47.8 | |
| Fiction.LiveBench | knowledge | 46.9 | |
| WeirdML | coding | 46.1 | |
| OSWorld | agentic | 43.9 | |
| ARC-AGI | reasoning | 40.0 | |
| GeoBench | knowledge | 37.0 | |
| Cybench | coding | 35.0 | |
| SimpleBench | reasoning | 34.6 | |
| The Agent Company | agentic | 33.1 | |
| ARC-AGI-2 | reasoning | 5.9 | |
| GSO-Bench | coding | 4.9 | |
| FrontierMath-2025-02-28-Private | math | 4.1 | |
| HLE | knowledge | 3.1 | |
| VPCT | knowledge | 1.0 | |
| FrontierMath-Tier-4-2025-07-01-Private | math | 0.1 |