Compare / DeepSeek V3.1 Terminus vs gpt-oss-120b
DeepSeek V3.1 Terminus vs gpt-oss-120b
20 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | DeepSeek V3.1 Terminus | gpt-oss-120b |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,423 | 1,351 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,405 | 1,277 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,420 | 1,323 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,417 | 1,352 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,411 | 1,309 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 43.5 | 30.4 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 14.8 | 11.6 |
| Artificial Analysisreported | GPQA Diamond% of questions | 79.2% | 78.2% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 16.4% | 19.6% |
| Artificial Analysisreported | IFBench% of instructions | 57.0% | 69.0% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 69.3% | 52.0% |
| Artificial Analysisreported | SciCode% of problems | 38.0% | 34.0% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 44.9% | 26.2% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.0% | 0.0% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 31.8% | 23.5% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 21.0% | 12.8% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 37.1% | 65.8% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 32.0% | 107.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.37 | 0.29 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 54.0 | 38.5 |
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.