Qwen2.5 72B Instruct vs DeepSeek-V2 (MoE-236B, May 2024)
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Qwen2.5 72B Instruct | DeepSeek-V2 (MoE-236B, May 2024) |
|---|---|---|
| Provider |  Alibaba Qwen |  DeepSeek |
| pontuação média | 53.2 | 76.5 |
| Preço de entrada | $0.12 | - |
| Preço de saída | $0.39 | - |
| Janela de contexto | 33K tokens (~16 books) | - |
| Lançado em | 2024-09-19 | 2024-01-01 |
| Código aberto | Open Source | Open Source |
Pontuações de benchmark
7 benchmarks · Qwen2.5 72B Instruct: 3, DeepSeek-V2 (MoE-236B, May 2024): 4
| Benchmark | Categoria | Qwen2.5 72B Instruct | DeepSeek-V2 (MoE-236B, May 2024) |
|---|---|---|---|
| ARC AI2 | knowledge | 92.7 | 89.6 |
| BBH | reasoning | 73.1 | 71.7 |
| HellaSwag | knowledge | 79.7 | 82.8 |
| MMLU | knowledge | 80.4 | 71.2 |
| PIQA | knowledge | 65.2 | 67.8 |
| TriviaQA | knowledge | 71.9 | 80.0 |
| Winogrande | knowledge | 64.6 | 72.6 |