Claude Opus 4.8 vs Claude Opus 5
| model Claude Opus 4.8 Anthropic | model Claude Opus 5 Anthropic | |
|---|---|---|
| Context | 1M | 1M |
| Input $/M | — | — |
| Output $/M | — | — |
| Knowledge cutoff | 2026-01 | 2026-05 |
| Available via | Anthropic API | Anthropic API |
| SWE-bench Pro | 69.2% | 79.2% |
| Terminal-Bench (success rate) | 74.6 | — |
| SWE-bench Verified (resolved) | 88.6 | 96 |
| Humanity's Last Exam (accuracy) | 49.8 | — |
| OSWorld-Verified (success rate) | 83.4 | — |
| SWE-bench Multilingual | — | 89.5% |
| SWE-bench Multimodal | — | 59.4% |
| DeepSWE | — | 68.8% |
| FrontierCode (mean@5) | — | 53.4 |