So is gpt-5-codex-high better than gpt-5-high? I find it hard to believe that with half the instruction tokens a MOE model could significantly outperform, but I

So is gpt-5-codex-high better than gpt-5-high? I find it hard to believe that with half the instruction tokens a MOE model could significantly outperform, but I am certainly not knowledgable in fine-tuning.