GPT-5.2 and GPT-5.2-Codex now run about 40% faster, weights unchanged

We have optimized our inference stack for API customers and GPT-5.2 and GPT-5.2-Codex now run ~40% faster. Model and model weights are unchanged.