Qwen2.5-Math-72B: benchmark results
Alibaba's Qwen2.5-Math-72B, a math-specialized base model fine-tuned from Qwen2.5-72B, supporting Chain-of-Thought and Tool-integrated Reasoning for English and Chinese math problems. Provider: Alibaba. Released 2024-09-19. Access: Open.
Aggregate Rating 1516 ± 37, rank #727 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PhysicsFinals | 32.2 | Score (imported) | 81.2 |
| MERA - SimpleAr | 99.6 | EM (%) | 80 |
| MERA - ruModAr | 83.17 | EM (%) | 77.3 |
| MERA - ruMultiAr | 44.14 | EM (%) | 76.7 |
| MERA - MathLogicQA | 73.23 | Accuracy (%) | 75.7 |
| MERA - BPS | 98.4 | Accuracy (%) | 75.2 |
| MERA - PARus | 88.4 | Accuracy (%) | 68.7 |
| MERA - ruMMLU | 67.34 | Accuracy (%) | 62.9 |
| MERA - ruHateSpeech | 77.74 | Accuracy (%) | 58.3 |
| MERA - MaMuRAMu | 74.58 | Accuracy (%) | 56.9 |
| MERA - ruOpenBookQA | 86.25 | Accuracy (%) | 55.7 |
| MERA | 0.51 | Total score | 55 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-math-72b · How It Works · Data refreshed daily, snapshot 2026-10-20.