As you might've noticed above, I've had access to a version of GPT-5 early. It sets the new SoTA by a significant margin on this benchmark and does much better than o3-high. It's a great model. On the other hand, Anthropic's best model lags. Google's is middle of the pack.