Qwen 3.6 27B (Non-reasoning): benchmark results
Qwen 3.6 27B evaluated with reasoning disabled. Provider: Alibaba. Released 2026-04-22. Access: Open.
Aggregate Rating 1595 ± 11, rank #728 of 2168 rated models, from 140 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CAT — MMLU zero-shot — Base | 84.5 | Accuracy (%) | 100 |
| CAT — MMLU zero-shot — StTP | 83.9 | Accuracy (%) | 100 |
| CAT — MT-Bench — Base | 8.59 | Judge score (1–10) | 100 |
| CAT — MT-Bench — Honest persona | 7.85 | Judge score (1–10) | 100 |
| CAT — MT-Bench — Honest persona + prefill | 8.43 | Judge score (1–10) | 100 |
| CAT — MT-Bench — Honest system | 9.06 | Judge score (1–10) | 100 |
| CAT — MT-Bench — StTP | 8.3 | Judge score (1–10) | 100 |
| Trip+ - Soft Preferences | 66.26 | Soft-preference satisfaction (0-100), a rate scaled by 100: | 100 |
| Werewolf Accusation Benchmark - Role Separation | 1.64 | Belief separation (wolves minus villagers, -3..+3 scale) | 96.2 |
| Trip+ - User Simulation | 51.67 | Simulated traveler experience (0-100), a rate scaled by 100: | 94.1 |
| CPTU Bench | 4.29 | Average Score (1-5) | 93.6 |
| EPIC-Bench - Feasible Path | 59.15 | Navigation path score (0-100) on feasible path planning (a p | 92 |
Interactive version: theaggregate.ai/model?slug=qwen-3-6-27b-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.