Compare / GPT-5.6 Luna vs gpt-oss-120b
GPT-5.6 Luna vs gpt-oss-120b
20 results from 4 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.6 Luna | gpt-oss-120b |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,461 | 1,385 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,463 | 1,351 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,412 | 1,277 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,491 | 1,357 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,446 | 1,323 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,454 | 1,352 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,435 | 1,309 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 71.4 | 30.4 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 37.3 | 11.6 |
| Artificial Analysisreported | GPQA Diamond% of questions | 91.1% | 78.2% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 39.5% | 19.6% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 83.7% | 52.0% |
| Artificial Analysisreported | SciCode% of problems | 53.6% | 34.0% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 80.9% | 26.2% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 11.6% | 0.0% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 31.1% | 12.8% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 15.0% | 107.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.38 | 0.29 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 53.7 | 38.5 |
| Vending-Bench 2reported | Money after a yearUS dollars | $4,094.71 | $-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.