Compare / Claude Sonnet 4.5 vs GPT-5.2
Claude Sonnet 4.5 vs GPT-5.2
34 results from 6 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Claude Sonnet 4.5 | GPT-5.2 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,449 | 1,425 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,175 | 1,162 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,449 | 1,442 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,461 | 1,444 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,454 | 1,391 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,498 | 1,475 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,464 | 1,427 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,457 | 1,437 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,454 | 1,411 |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 95.0% | 87.3% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 65.0% | 45.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 61.4% | 43.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 73.2% | 55.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 68.8% | 75.0% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 88.7% | 81.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 81.1% | 70.4% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 81.0% | 75.5% |
| OpenHands Indexreported | Average% resolved | 53.0% | 58.8% |
| OpenHands Indexreported | Front end (SWE-Bench Multimodal)% resolved | 36.8% | 30.9% |
| OpenHands Indexreported | Greenfield (Commit0)% resolved | 12.5% | 50.0% |
| OpenHands Indexreported | Information gathering (GAIA)% resolved | 72.7% | 65.5% |
| OpenHands Indexreported | Issue resolution (SWE-Bench)% resolved | 74.2% | 74.6% |
| OpenHands Indexreported | Testing (SWT-Bench)% resolved | 68.8% | 73.2% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 48.0% | 72.0% |
| tau2-benchreported | Task success (pass^1)% of tasks | 72.0% | 83.0% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 39.5% | 51.8% |
| tau2-benchreported | Task success (pass^1)% of tasks | 72.4% | 81.6% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 64.0% | 71.9% |
| tau2-benchreported | Task success (pass^1)% of tasks | 84.9% | 89.7% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 5.0% | 76.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.41 | 0.40 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 66.9 | 49.4 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 95.6% | 100.0% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 12.0% | 8.4% |
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.