Compare / Claude Opus 4.6 vs Claude Sonnet 4.5
Claude Opus 4.6 vs Claude Sonnet 4.5
20 results from 4 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Claude Opus 4.6 | Claude Sonnet 4.5 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,495 | 1,449 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,223 | 1,175 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,494 | 1,449 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,501 | 1,461 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,501 | 1,454 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,547 | 1,498 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,513 | 1,464 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,505 | 1,457 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,500 | 1,454 |
| OpenHands Indexreported | Average% resolved | 66.7% | 53.0% |
| OpenHands Indexreported | Front end (SWE-Bench Multimodal)% resolved | 41.8% | 36.8% |
| OpenHands Indexreported | Greenfield (Commit0)% resolved | 56.2% | 12.5% |
| OpenHands Indexreported | Information gathering (GAIA)% resolved | 80.0% | 72.7% |
| OpenHands Indexreported | Issue resolution (SWE-Bench)% resolved | 76.8% | 74.2% |
| OpenHands Indexreported | Testing (SWT-Bench)% resolved | 78.8% | 68.8% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 3.0% | 5.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.41 | 0.41 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 70.9 | 66.9 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 99.8% | 95.6% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 12.2% | 12.0% |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.