Claude Sonnet 4.5 (Thinking 32K): benchmark results
Claude Sonnet 4.5 evaluated with a 32K-token thinking budget. Provider: Anthropic. Released 2025-09-29. Access: API.
Aggregate Rating 1686 ± 24, rank #405 of 2168 rated models, from 22 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MATH Level 5 | 97.73 | Accuracy (%) | 95.3 |
| Korean CSAT 2026 (Easy Mode) - Mathematics | 100 | Points (out of 100) | 88.5 |
| Korean CSAT 2026 (Easy Mode) - Korean | 97.5 | Points (out of 100) | 74.7 |
| Korean CSAT 2026 (Easy Mode) - Society and Culture | 43 | Points (out of 50) | 74 |
| Korean CSAT 2026 (Easy Mode) - Total | 422 | Points (out of 450) | 68.2 |
| Korean CSAT 2026 (Easy Mode) - English | 97 | Points (out of 100) | 66.2 |
| Epoch AI - GPQA Diamond | 81.73 | Accuracy (%) | 61.3 |
| OTIS Mock AIME 2024-25 | 77.78 | Accuracy (%) | 58.1 |
| Korean CSAT 2026 (Easy Mode) - Life Science I | 39 | Points (out of 50) | 56.8 |
| Korean CSAT 2026 (Easy Mode) - Chemistry I | 43 | Points (out of 50) | 54.1 |
| FrontierMath - Tiers 1-3 | 15.22 | Accuracy (%, 290 problems) | 51.5 |
| Korean CSAT 2026 (Easy Mode) - Physics I | 30 | Points (out of 50) | 50 |
Interactive version: theaggregate.ai/model?slug=claude-sonnet-4-5-thinking-32k · How It Works · Data refreshed daily, snapshot 2026-10-20.