Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance

Benchmark Model Rank Results
instruction-following-on-mtbenchFalcon-H1-34B-Instruct#2Score (1-10): 9.20
multi-task-language-understanding-on-mgsmFalcon-H1-34B-Instruct (8-shot)#7Average (%): 76.33
multi-task-language-understanding-on-mmluFalcon-H1-34B-Instruct (5-shot)#10Average (%): 84.05
code-generation-on-mbppFalcon-H1-34B-Instruct#21Accuracy: 83.86
code-generation-on-humanevalFalcon-H1-34B-Instruct#39Pass@1: 87.2
code-generation-on-livecodebench-v6Falcon-H1-34B-Instruct#50Pass@1: 49.71
sentence-completion-on-hellaswagFalcon-H1-34B-Instruct#61Accuracy: 81.94
mathematical-reasoning-on-math-500Falcon-H1-34B-Instruct#78Accuracy: 83.80
multi-task-language-understanding-on-mmlu-proFalcon-H1-34B-Instruct (5-shot)#99Accuracy: 58.73
mathematical-reasoning-on-aime24Falcon-H1-34B-Instruct#112Acc: 23.75
multiple-choice-question-answering-mcqa-on-gpqa-diamondFalcon-H1-34B-Instruct#112Accuracy: 49.66
arithmetic-reasoning-on-gsm8kFalcon-H1-34B-Instruct (5-shot)#128Accuracy: 83.62
mathematical-reasoning-on-aime-2025Falcon-H1-34B-Instruct#133Accuracy: 16.67