How does this compare with existing frontier models especially when processing massive volumes of evidence that might be contradictory...