Compare / Gemini 2.5 Flash vs Gemma 3 27B
Gemini 2.5 Flash vs Gemma 3 27B
17 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Gemini 2.5 Flash | Gemma 3 27B |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,437 | 1,380 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,399 | 1,377 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,394 | 1,348 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,424 | 1,337 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,400 | 1,343 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,409 | 1,365 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,402 | 1,354 |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 93.7% | 73.7% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 41.3% | 13.6% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 36.2% | 10.8% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 56.2% | 29.5% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 75.0% | 81.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 88.1% | 87.2% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 78.2% | 74.5% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 62.0% | 0.0% |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 99.0% | 98.8% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 7.8% | 7.4% |
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.