Compare / Qwen3.5 397B A17B vs GPT-6 Astra
Qwen3.5 397B A17B vs GPT-6 Astra
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 | Qwen3.5 397B A17B | GPT-6 Astra |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 24.9% | 64.7% |
| Arena (formerly LMArena)reported | OverallArena rating | 1,445 | 1,465 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,445 | 1,483 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,405 | 1,453 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,476 | 1,510 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,434 | 1,473 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,442 | 1,478 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,421 | 1,469 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 48.2 | 77.1 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 21.4 | 52.7 |
| Artificial Analysisreported | GPQA Diamond% of questions | 89.3% | 96.3% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 29.0% | 54.7% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 77.3% | 80.7% |
| Artificial Analysisreported | SciCode% of problems | 44.8% | 56.5% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 51.3% | 89.9% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.0% | 59.6% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 13.4% | 43.1% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 23.0% | 0.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.33 | 0.38 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 50.1 | 68.9 |
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.