OpenAI uses Codex to keep ChatGPT fast OpenAI cut ChatGPT's p99 latency by roughly 30% using an automated, Codex-driven "performance factory" that continuously profiles, patches and redeploys production code, the company said in a post on its website. OpenAI said the system runs 24/7 so that each new model version or dependency update receives the same treatment, reducing surprise regressions and manual tuning. The company framed the result as allowing larger models or denser prompts without hitting the same latency wall, or keeping the same SLOs while cutting 20–30% off the inference cloud bill. OpenAI https://openai.com/index/how-openai-performance-factory-keeps-chatgpt-fast OpenAI uses Codex to keep ChatGPT fast Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. OpenAI cut ChatGPT’s p99 latency by ~30% using an automated, Codex-driven performance factory that continuously profiles, patches, and redeploys production code. This means you can now ship larger models or denser prompts without hitting the same latency wall, or you can keep the same SLOs while shaving 20–30% off your inference cloud bill. The factory runs 24/7, so every new model version or dependency update gets the same treatment—expect fewer surprise regressions and less manual tuning. OpenAI's Performance Factory leverages Codex-powered workflows to automatically detect and resolve performance bottlenecks in ChatGPT, reducing latency and improving response times. This ensures more reliable and scalable deployment of LLMs in production environments.