Qwen2.5-72B-Instruct-AWQ: benchmark results
Provider: Alibaba. Released 2024-09-19. Access: Open.
Aggregate Rating 1573 ± 36, rank #1361 of 10916 rated models, from 13 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| VMLU | 69.17 | Average (%) | 95.8 |
| VMLU - Other | 64.62 | Accuracy (%) | 95.8 |
| VMLU - STEM | 68.46 | Accuracy (%) | 95.8 |
| VMLU - Humanities | 69.23 | Accuracy (%) | 91.7 |
| VMLU - Social Science | 74.65 | Accuracy (%) | 91.7 |
| Phun-Bench - Contextual Homophone Recognition | 53 | Recognition rate (%): share of homophonic pun sentences (1,6 | 50 |
| Phun-Bench - Rhyming Sentence Generation (1-Rhyme) | 55.5 | Exact-match rate (%) of the rhyme group of the generated sen | 42.9 |
| Phun-Bench - Rhyming Sentence Generation (2-Rhyme) | 5.3 | Exact-match rate (%) of the rhyme groups of the generated se | 42.9 |
| Phun-Bench - Homophone Recall | 35.1 | Accuracy (%) of recovering a four-character chengyu from an | 37.5 |
| Multilingual Judges - SSA-MTE Cross-Ling. | 0.34 | Spearman correlation | 33.3 |
| Multilingual Judges - SSA-MTE Monolingual | 0.36 | Spearman correlation | 33.3 |
| Multilingual Judges - SSA-MTE Trans-Eval | 0.25 | Spearman correlation | 33.3 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-72b-instruct-awq · How It Works · Data refreshed daily, snapshot 2026-11-08.