Compare / Qwen3 235B A22B Thinking 2507 vs DeepSeek V4 Flash 0423
Qwen3 235B A22B Thinking 2507 vs DeepSeek V4 Flash 0423
18 results from 3 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | Qwen3 235B A22B Thinking 2507 | DeepSeek V4 Flash 0423 |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,404 | 1,439 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,370 | 1,408 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,466 | 1,457 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,385 | 1,428 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,400 | 1,436 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,377 | 1,416 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 22.1 | 69.1 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 12.7 | 34.3 |
| Artificial Analysisreported | GPQA Diamond% of questions | 79.0% | 90.8% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 15.9% | 38.6% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 72.0% | 79.7% |
| Artificial Analysisreported | SciCode% of problems | 41.4% | 50.3% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 12.0% | 78.7% |
| Artificial Analysisreported | Terminal-Bench 4.0% of tasks | 0.0% | 12.1% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 7.8% | 39.4% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 30.0% | 12.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.31 | 0.36 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 32.4 | 54.6 |
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.