Qwen2.5 72B Instruct
Qwen · released 2024-09-19 · other license
Qwen2.5-72B-Instruct is Alibaba's flagship 2024 open-weight non-reasoning model: vendor GPQA 49.0, MMLU-Pro 71.1, LiveCodeBench 55.5. A leading open model of its time.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 72.71B total |
| Architecture | qwen2 |
| Context window | 131K tokens |
| Knowledge cutoff | 2024-06-30 |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | 49% |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | 71.1% |
| BFCL v3 (tool use) | — |
| Composite score | 6.67 |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 43.7 GB | 42.2 GB | — | 131K |
Strengths & weaknesses
Strengths: Strong math and reasoning (vendor MATH 83.1); Leading multilingual instruction following; Top open intelligence in its 2024 size class
Weaknesses: No native reasoning mode; Restrictive (non-Apache) Qwen license at 72B