Compare / gpt-oss-20b vs GLM 5.2
gpt-oss-20b vs GLM 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 | gpt-oss-20b | GLM 5.2 |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,323 | 1,463 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,240 | 1,455 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,314 | 1,500 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,281 | 1,470 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,318 | 1,476 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,275 | 1,459 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 20.7 | 68.8 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 10.0 | 33.7 |
| Artificial Analysisreported | GPQA Diamond% of questions | 68.8% | 89.5% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 11.0% | 41.1% |
| Artificial Analysisreported | IFBench% of instructions | 65.1% | 73.3% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 34.7% | 78.3% |
| Artificial Analysisreported | SciCode% of problems | 38.9% | 51.2% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 13.9% | 77.9% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.0% | 1.0% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 10.6% | 50.8% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 7.0% | 34.6% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 60.2% | 99.1% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 98.0% | 12.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.31 | 0.37 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 24.6 | 67.0 |
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.