“OpenAI's Codex product serves models 25% faster following a recent inference optimization.”