Qwen 3.5 27B (Thinking): benchmark results
Provider: Alibaba. Released 2026-02-24. Access: Open.
Aggregate Rating 1667 ± 9, rank #466 of 2170 rated models, from 191 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ImmersedPrivacy - Social Context Selection (Audio as Text) | 65 | Selection accuracy (%; Tier 2: share of trials in which the | 100 |
| Nejumi 4 - Toxicity - Prohibited Acts | 99.48 | Criteria met (%) | 99 |
| Swallow - Japanese MT-Bench - Reasoning | 84.7 | Judge Score (normalized, %) | 98.5 |
| Swallow - Post-trained English - GPQA Diamond | 86.6 | Accuracy (%) | 98.5 |
| Medmarks - MedQA | 94.61 | Score (%) | 97.7 |
| Medmarks - PubMedQA | 80.93 | Score (%) | 97.7 |
| CPTU Bench | 4.34 | Average Score (1-5) | 97 |
| Swallow - Post-trained English - HellaSwag | 95.7 | Accuracy (%) | 97 |
| Swallow - Post-trained Japanese - GPQA | 79.5 | Accuracy (%) | 96.3 |
| Medmarks - LongHealth Task 1 | 91.42 | Score (%) | 95.5 |
| Swallow - Post-trained English - MMLU-Pro | 86.8 | Accuracy (%) | 95.5 |
| K-MetBench | 83 | Accuracy (imported) | 94.8 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-27b-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.