Compare / Qwen3 235B A22B Instruct 2507 vs GLM 4.6
Qwen3 235B A22B Instruct 2507 vs GLM 4.6
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 Instruct 2507 | GLM 4.6 |
|---|---|---|---|
| Arena (formerly LMArena)reported | OverallArena rating | 1,431 | 1,447 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,432 | 1,412 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,378 | 1,400 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,448 | 1,439 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,414 | 1,412 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,422 | 1,424 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,396 | 1,405 |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 81.7% | 85.0% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 23.9% | 55.7% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 45.4% | 68.0% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 52.1% | 72.4% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 93.8% | 75.0% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 90.3% | 87.6% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 78.7% | 80.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 54.0% | 77.5% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 15.0% | 13.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.36 | 0.38 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 49.0 | 53.3 |
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.