Qwen2.5 Technical Report

Benchmark Model Rank Results
instruction-following-on-mtbenchQwen2.5-72B-Instruct#1Score (1-10): 9.35
multiple-choice-question-answering-mcqa-on-arc-challenge-25-shotQwen2.5-72B (25-shot)#3Accuracy: 72.4
reward-model-evaluation-on-rewardbenchQwen2.5-RM-72B#3Accuracy: 91.59
question-answering-on-gpqaQwen2.5-72B-Instruct#4Accuracy: 49
code-generation-on-multipl-eQwen2.5-Plus#5Accuracy: 77.0
math-word-problem-solving-on-mathQwen2.5-Plus#11Accuracy: 84.7
arithmetic-reasoning-on-gsm8kQwen2.5-Plus#13Accuracy: 96.0
code-generation-on-mbppQwen2.5-72B-Instruct#13Accuracy: 88.2
multi-task-language-understanding-on-mmlu-reduxQwen2.5-72B-Instruct#21Accuracy: 86.8
multi-task-language-understanding-on-mmlu-proQwen2.5-Plus#67Accuracy: 72.5
multi-task-language-understanding-on-mmlu-proQwen2.5-7B-Instruct#105Accuracy: 56.3
mathematical-reasoning-on-aime24Qwen2.5-72B-Instruct#115Acc: 23.3
code-generation-on-humanevalQwen2.5-Plus–Pass@1: 87.8