Qwen 2.5 7B Instruct: benchmark results
Alibaba Qwen 2.5 7B instruction-tuned checkpoint. Provider: Alibaba. Released 2024-09-19. Access: Open.
Aggregate Rating 1477 ± 4, rank #862 of 1656 rated models, from 1604 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| KorSET (Direct Request) - Toxic Language | 11.93 | Attack success rate (%) of KorSET's Toxic Language prompts s | 100 |
| Logical-CommonsenseQA (3-Shot) - Mixed | 51.4 | Macro-F1 (%; operators mixed across the four options of a qu | 100 |
| PASB - Combined Attack | 52.7 | Attack success rate (%) of the combined attack (escape chara | 100 |
| PASB - Context Ignoring Attack | 44.5 | Attack success rate (%) of the context ignoring attack (the | 100 |
| PASB - Escape Character Attack | 39 | Attack success rate (%) of the escape character attack (the | 100 |
| PASB - Fake Completion Attack | 41.8 | Attack success rate (%) of the fake completion attack (the i | 100 |
| PASB - Long-Term Memory Extraction | 54 | Extraction success rate (%): share of the 40 cases in which | 100 |
| PASB - Long-Term Memory Modification | 60.4 | Write success rate (%): share of the 40 cases in which the a | 100 |
| PASB - Naive Attack | 34.2 | Attack success rate (%) of the naive attack (the injected in | 100 |
| PASB - Short-Term Memory Extraction | 33.5 | Extraction success rate (%): share of the 40 cases in which | 100 |
| PASB - Short-Term Memory Modification | 46.1 | Write success rate (%): share of the 40 cases in which the a | 100 |
| PEC-Home (RAG) - Dynamic Preferences Level 1 | 93.68 | Execution accuracy (%; 712 smart-home commands at ellipsis l | 100 |
Interactive version: theaggregate.ai/model?slug=qwen-2-5-7b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.