Compare / Qwen3 235B A22B Instruct 2507 vs qwen3-4b-instruct-2507
Qwen3 235B A22B Instruct 2507 vs qwen3-4b-instruct-2507
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 | qwen3-4b-instruct-2507 |
|---|---|---|---|
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 12.0 | 6.70 |
| Artificial Analysisreported | GPQA Diamond% of questions | 75.3% | 51.7% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 11.1% | 4.5% |
| Artificial Analysisreported | IFBench% of instructions | 46.1% | 33.5% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 33.9% | 11.3% |
| Artificial Analysisreported | Terminal-Bench Hard% of tasks | 15.2% | 4.5% |
| Artificial Analysisreported | Τ²-bench telecom% of tasks | 33.3% | 26.6% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Irrelevance detection% correct | 81.7% | 84.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Memory% correct | 23.9% | 23.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Multi-turn tasks% correct | 45.4% | 22.1% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Overall accuracy% correct | 52.1% | 35.7% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Relevance detection% correct | 93.8% | 87.5% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (curated)% correct | 90.3% | 87.9% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Single-turn calls (user-contributed)% correct | 78.7% | 76.4% |
| Berkeley Function Calling Leaderboard (BFCL) V4reported | Web search% correct | 54.0% | 4.5% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 15.0% | 8.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.36 | 0.34 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 49.0 | 29.9 |
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.