Artificial Analysis just benchmarked them and the scores are crazy good, proving the earlier success wasn't only enabled by overthinking.
