Compare / Phi 4 vs o4 Mini
Phi 4 vs o4 Mini
19 results from 4 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Phi 4 | o4 Mini |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,256 | 1,379 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,210 | 1,337 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,268 | 1,408 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,245 | 1,369 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,256 | 1,391 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,229 | 1,354 |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 87.5% | 87.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 24.7% | 35.3% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 3.9% | 41.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 28.8% | 53.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 50.0% | 81.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 69.6% | 81.3% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 60.7% | 70.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 4.5% | 75.5% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 19.0% | 19.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.34 | 0.33 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 25.7 | 46.9 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 80.7% | 98.7% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 3.7% | 18.6% |
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.