Compare / GPT-5.2 vs o3
GPT-5.2 vs o3
19 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.2 | o3 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,162 | 1,192 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,442 | 1,405 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,444 | 1,426 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,391 | 1,383 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,475 | 1,449 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,427 | 1,403 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,437 | 1,432 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,411 | 1,396 |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 87.3% | 86.1% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 45.8% | 51.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 43.8% | 62.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 55.9% | 63.0% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 75.0% | 93.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 81.8% | 81.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 70.4% | 73.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 75.5% | 77.0% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 76.0% | 18.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.40 | 0.30 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 49.4 | 66.8 |
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.