Qwen-3-Swallow-32B-SFT-v0.2: benchmark results
Provider: Alibaba.
Aggregate Rating 1601 ± 32, rank #1001 of 10916 rated models, from 35 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Swallow - English MT-Bench - Extraction | 83 | Judge Score (normalized, %) | 91.8 |
| Swallow - English MT-Bench - Writing | 74.3 | Judge Score (normalized, %) | 72.4 |
| Swallow - Japanese MT-Bench - Coding | 78.7 | Judge Score (normalized, %) | 71.6 |
| Swallow - Japanese MT-Bench - Roleplay | 69.3 | Judge Score (normalized, %) | 71.6 |
| Swallow - Japanese MT-Bench - Average | 73.8 | Judge Score (normalized, %) | 68.7 |
| Swallow - Post-trained English - IFBench | 47.7 | Instruction-Level Strict Accuracy (%) | 67.2 |
| Swallow - Japanese MT-Bench - Math | 97.5 | Judge Score (normalized, %) | 66.4 |
| Swallow - Post-trained Japanese - WMT20 En-Ja | 24.7 | BLEU | 66.4 |
| Swallow - Japanese MT-Bench - Writing | 65.8 | Judge Score (normalized, %) | 65.7 |
| Swallow - Post-trained Japanese - GPQA | 60.9 | Accuracy (%) | 64.9 |
| Swallow - Post-trained Japanese - JHumanEval | 91.1 | Pass@1 (%) | 64.9 |
| Swallow - Japanese MT-Bench - STEM | 66.5 | Judge Score (normalized, %) | 62.7 |
Interactive version: theaggregate.ai/model?slug=qwen-3-swallow-32b-sft-v0-2 · How It Works · Data refreshed daily, snapshot 2026-11-08.