Qwen2.5 72B Instruct
Código abertopor Alibaba Qwen · Lançado em 2024-09-19
52.3
pontuação média
$0.12/1M
Preço de entrada
$0.39/1M
Preço de saída
33K tokens (~16 books)
Janela de contexto
text
Tipo
Tested on 15 benchmarks with 52.3% average. Top scores: ARC AI2 (92.7%), MMLU (80.4%), HellaSwag (79.7%).
Pontuações de benchmark
| Benchmark | Categoria | Pontuação | Bar |
|---|---|---|---|
| ARC AI2 | knowledge | 92.7 | |
| MMLU | knowledge | 80.4 | |
| HellaSwag | knowledge | 79.7 | |
| BBH | reasoning | 73.1 | |
| TriviaQA | knowledge | 71.9 | |
| PIQA | knowledge | 65.2 | |
| VideoMME | multimodal | 64.7 | |
| Winogrande | knowledge | 64.6 | |
| MATH level 5 | math | 63.2 | |
| GeoBench | knowledge | 62.0 | |
| GPQA diamond | knowledge | 32.2 | |
| Balrog | knowledge | 16.2 | |
| OTIS Mock AIME 2024-2025 | math | 8.0 | |
| The Agent Company | agentic | 5.7 | |
| OSWorld | agentic | 5.0 |