Compare / Gemini 3.8 Flash vs gpt-oss-120b
Gemini 3.8 Flash vs gpt-oss-120b
21 results from 5 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Gemini 3.8 Flash | gpt-oss-120b |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 64.3% | 4.4% |
| Arena (formerly LMArena)reported | OverallArena rating | 1,477 | 1,385 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,482 | 1,351 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,482 | 1,277 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,529 | 1,357 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,488 | 1,323 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,492 | 1,352 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,486 | 1,309 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 76.3 | 30.4 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 40.9 | 11.6 |
| Artificial Analysisreported | GPQA Diamond% of questions | 95.3% | 78.2% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 47.8% | 19.6% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 84.0% | 52.0% |
| Artificial Analysisreported | SciCode% of problems | 56.6% | 34.0% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 87.6% | 26.2% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 19.7% | 0.0% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 45.8% | 12.8% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 13.0% | 107.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.34 | 0.29 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 78.5 | 38.5 |
| Vending-Bench 2reported | Money after a yearUS dollars | $5,093.79 | $-21.53 |
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.