Compare / gemini-3-flash vs GPT-5.2
gemini-3-flash vs GPT-5.2
21 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-3-flash | GPT-5.2 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,439 | 1,425 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,191 | 1,162 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,468 | 1,442 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,469 | 1,444 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,457 | 1,391 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,496 | 1,475 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,458 | 1,427 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,473 | 1,437 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,461 | 1,411 |
| OpenHands Indexreported | Average% resolved | 49.0% | 58.8% |
| OpenHands Indexreported | Front end (SWE-Bench Multimodal)% resolved | 22.1% | 30.9% |
| OpenHands Indexreported | Greenfield (Commit0)% resolved | 18.8% | 50.0% |
| OpenHands Indexreported | Information gathering (GAIA)% resolved | 58.8% | 65.5% |
| OpenHands Indexreported | Issue resolution (SWE-Bench)% resolved | 74.6% | 74.6% |
| OpenHands Indexreported | Testing (SWT-Bench)% resolved | 70.7% | 73.2% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 68.0% | 72.0% |
| tau2-benchreported | Task success (pass^1)% of tasks | 82.5% | 83.0% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 51.8% | 51.8% |
| tau2-benchreported | Task success (pass^1)% of tasks | 76.8% | 81.6% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 70.2% | 71.9% |
| tau2-benchreported | Task success (pass^1)% of tasks | 91.2% | 89.7% |
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.