Qwen 3.5 27B: benchmark results
Alibaba's dense open Qwen 3.5 27B multimodal model. Provider: Alibaba. Released 2026-02-24. Access: Open.
Aggregate Rating 1704 ± 3, rank #187 of 1656 rated models, from 1814 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AIME-TR | 76.67 | Accuracy (%, AIME 2024+2025 in Turkish) | 100 |
| ArtECulture - Arabic | 41.53 | Accuracy (%; Arabic-speaking culture labels; zero-shot predi | 100 |
| BARISTA - Activity Recognition | 77.8 | Accuracy (%) of 25-way multiple-choice activity recognition | 100 |
| BARISTA - Temporal VQA | 0.7 | Mean of answer accuracy and completeness, Gemini 2.5 Flash G | 100 |
| Cloud-OpsBench (OnlineBoutique) - Evidence-Order Consistency | 70 | Evidence-order consistency (%): mean share of milestone grou | 100 |
| Cloud-OpsBench (OnlineBoutique) - Milestone Coverage | 75 | Milestone coverage (%): mean share of the case's expert diag | 100 |
| DistortBench | 61.9 | Joint accuracy (%) on DistortBench's 13,500 four-choice ques | 100 |
| DistortBench - Color | 58.3 | Joint accuracy (%) on the five color distortions (diffusion, | 100 |
| DistortBench - Luminance | 59.2 | Joint accuracy (%) on the four luminance distortions (bright | 100 |
| DistortBench - Noise | 65.7 | Joint accuracy (%) on the five noise distortions (white, col | 100 |
| DistortBench - Spatial | 64.8 | Joint accuracy (%) on the eight spatial distortions (jitter, | 100 |
| EYT-Bench (Nemotron-USA) - Empathy | 77.4 | Empathy (recognition, attribution, resonance, response, supp | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-27b · How It Works · Data refreshed daily, snapshot 2026-10-20.