Llama 3.1 405B vs Qwen2.5 72B Instruct
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Llama 3.1 405B | Qwen2.5 72B Instruct |
|---|---|---|
| Provider |  Meta |  Alibaba Qwen |
| puntuación promedio | 38.0 | 53.2 |
| Precio de entrada | - | $0.12 |
| Precio de salida | - | $0.39 |
| Ventana de contexto | - | 33K tokens (~16 books) |
| Publicado el | 2024-07-16 | 2024-09-19 |
| Código abierto | Open Source | Open Source |
Puntuaciones de benchmark
17 benchmarks · Llama 3.1 405B: 9, Qwen2.5 72B Instruct: 8
| Benchmark | Categoría | Llama 3.1 405B | Qwen2.5 72B Instruct |
|---|---|---|---|
| ARC AI2 | knowledge | 93.7 | 92.7 |
| BBH | reasoning | 77.2 | 73.1 |
| BBH (HuggingFace) | general | 7.8 | 61.9 |
| GPQA | knowledge | 5.9 | 16.7 |
| GPQA diamond | knowledge | 34.5 | 32.2 |
| HellaSwag | knowledge | 85.6 | 79.7 |
| IFEval | language | 18.1 | 86.4 |
| MATH level 5 | math | 49.8 | 63.2 |
| MATH Level 5 | math | 0.0 | 59.8 |
| MMLU | knowledge | 79.3 | 80.4 |
| MMLU-PRO | knowledge | 25.7 | 51.4 |
| MUSR | reasoning | 2.2 | 11.7 |
| OTIS Mock AIME 2024-2025 | math | 9.6 | 8.0 |
| PIQA | knowledge | 71.8 | 65.2 |
| The Agent Company | agentic | 7.4 | 5.7 |
| TriviaQA | knowledge | 82.7 | 71.9 |
| Winogrande | knowledge | 78.4 | 64.6 |