OpenAI is about to make long ChatGPT conversations a lot less painful to revisit. The company says it's rolling out an optimization next week that will dramatically speed up loading for giant chat threads in both ChatGPT and Codex.
In internal testing, a massive 741-turn conversation — weighing in at 231MB — went from an average load time of 27.62 seconds down to 1.66 seconds. That's a roughly 94% improvement.
The change targets how the client loads chat history, not how fast the model generates responses. OpenAI says the number of conversation entries that needed to be loaded dropped from 15,529 to 64, and the number of requests fell from 894 to just 16. Memory growth in the app also came down 41.2%.
For anyone who's been using ChatGPT or Codex long enough to build up a bloated thread, this should mean noticeably less waiting when you open an old conversation. Previously, the longer a thread got, the more the client struggled — because it had to chew through tons of historical messages. Now, OpenAI is simply cutting the amount of data the client has to load in the first place.
The details come from an OpenAI employee.