Alibaba (Qwen)
Qwen2.5-72B benchmark results
13 results on 10 benchmarks from 5 sources.
Also listed as: qwen2.5-72b-instruct · qwen2.5-vl-72b-instruct · Qwen2.5 72B Base · Qwen2.5 72b instruct · Qwen2.5 Instruct Turbo (72B) · Qwen2.5-VL-72B-Instruct · Qwen/Qwen2.5-72B-Instruct · Qwen/Qwen2.5-72B
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1303 | 283 | Overall, style control on; 39406 votes; 95% CI 1299–1307“qwen2.5-72b-instruct” | Official leaderboard | Oct 8, 2026 |
| LMArena Vision | 1121 | 123 | Overall, style control on; 3768 votes; 95% CI 1110–1132“qwen2.5-vl-72b-instruct” | Official leaderboard | Oct 2, 2026 |
| Open LLM Leaderboard v2 | 48.0% | 6 | Average of 6 normalised benchmarks; bfloat16; 72.706B params“Qwen/Qwen2.5-72B-Instruct” | Official leaderboard | Mar 13, 2025 |
| GPQA Diamond | 49.1% | 206 | Epoch AI run, accuracy | Independent run | Jan 27, 2025 |
| MMLU | 85.0% | 6 | Exact match; 5-shot | Paper / tech report via Epoch AI | Sep 19, 2024 |
| MMLU | 83.4% | 9 | Exact match; 5-shot | Official leaderboard via Epoch AI | Sep 19, 2024 |
| MMLU-Pro | 71.6% | 96 | Overall accuracy; data source: Self-Reported; 72B | Vendor-reported | Mar 11, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 8.1% | 218 | Epoch AI run, OTIS Mock AIME 2024–2025 | Independent run | Feb 25, 2025 |
| MATH Level 5 | 63.2% | 50 | Epoch AI run, accuracy | Independent run | Jan 27, 2025 |
| OSWorld (Verified) | 5.0% | 18 | OSWorld; agent: qwen2.5-vl-72b-instruct (100 steps) | Official leaderboard via Epoch AI | Jul 27, 2025 |
| OSWorld (Verified) | 4.4% | 19 | OSWorld; agent: qwen2.5-vl-72b-instruct (15 steps) | Official leaderboard via Epoch AI | Jul 27, 2025 |
| CMMLU | 85.7% | 8 | Five-shot average | Official leaderboard | Dec 6, 2024 |
| CMMLU | 84.7% | 6 | Zero-shot average | Official leaderboard | Dec 6, 2024 |
FAQ
Where do these Qwen2.5-72B scores come from?
From Epoch AI Benchmarking Hub, CMMLU leaderboard (GitHub README), LMArena leaderboard dataset, MMLU-Pro leaderboard (TIGER-Lab), and Hugging Face Open LLM Leaderboard v2 (archived). Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.