Qwen 2
#28 in Open-Source LLMsalibaba · v2 · since 2024-06-06 · 2× · last seen Jun 29, 2026
18
Momentum
Qwen 2 is a language model by Alibaba, available in a 1.5B variant. According to tests, the model achieved a Retrieval Accuracy of approximately 50.0%.
Momentum trend
04.04.03.07.
Features
| Benchmark Score (MMLU/Similar) | Qwen2-72B (Base): 84.2 on MMLU (5-shot); 37.9 on GPQA; 64.6 on HumanEval; 89.5 on GSM8K. Qwen2-7B (Base): 70.3 on MMLU. |
| Context Window | 32,768 tokens (Qwen2-1.5B Base); up to 128,000 tokens for Qwen2-7B-Instruct and Qwen2-72B-Instruct (with YARN) |
| Model Size (Parameters) | 1.5 billion parameters (Qwen2-1.5B); model family: 0.5B / 1.5B / 7B / 57B-A14B (MoE) / 72B |
| Price Tier | Open source / free (Apache 2.0 for 0.5B, 1.5B, 7B, 57B-A14B); Qwen2-72B under Qianwen License (custom commercial license). Weights freely available on Hugging Face and ModelScope. |
| Memory Requirement | ~4.63 GB VRAM in float16/bfloat16 format for Qwen2-1.5B (weights); Qwen2-72B requires ~144 GB VRAM in BF16 |