Comment by npn

from what I understand, it's because unlike the other models, MAI models haven't yet fine-tuned against the synthetic datasets specifically designed to boost the benchmark scores.