← comparisons / claude-opus-5-vs-claude-sonnet-5
Claude Opus 5 vs Claude Sonnet 5
01Head to head
| Claude Opus 5 | Claude Sonnet 5 | |
|---|---|---|
| Creator | Anthropic | Anthropic |
| Released | 2026-07-24 | 2026-06-30 |
| Context | 1M | 1M |
| Max output | 128K | 128K |
| Input $/M | — | — |
| Output $/M | — | — |
| Knowledge cutoff | 2026-05 | 2026-01 |
| Modalities | text + image → text | text + image → text |
| Capabilities | tool-use, extended-thinking | tool-use, extended-thinking |
| Providers | Anthropic API | Anthropic API |
02Benchmarks · vendor-reported unless verified
| Claude Opus 5 | Claude Sonnet 5 | |
|---|---|---|
| BrowseComp (accuracy) | — | 84.7 |
| DeepSWE | 68.8% | — |
| FrontierCode (mean@5) | 53.4 | — |
| OSWorld-Verified (success rate) | — | 81.2 |
| SWE-bench Multilingual | 89.5% | 78.3% |
| SWE-bench Multimodal | 59.4% | — |
| SWE-bench Pro | 79.2% | 63.2% |
| SWE-bench Verified (resolved) | 96 | 85.2 |
| Terminal-Bench (success rate) | — | 80.4 |