OpenAI brings back more legacy models, ponders «personality» for GPT-5

In an update on the GPT-5 rollout today, CEO Sam Altman launched a bevy of new features. Like a longer context window for GPT-5-thinking, and changes to the model picker.

The launch of GPT-5 has been a little bumpy, at best. The first reaction was from users missing GPT-4o, which was quickly returned — but what about the other «legacy models?»

Almost all models returned
They are all coming back, and as per now the model picker lists GPTs 4.1, o3 and o4-mini for paid users. The only one missing from before the GPT-5 launch is GPT-4.5, which Altman says «costs a lot of GPUs.»

Altman also says they are «working on an update» to GPT-5s personality, that «should feel warmer» — but not as annoying as many people felt 4o was.

— However, one learning for us from the past few days is we really just need to get to a world with more per-user customization of model personality, he says.

Personality matters
This makes it seem like the launch of GPT-5 was a huge wake-up call to OpenAI that personality matters almost as much as intelligence.

It looks like people were more satisfied with a bot they could crack open a cold one and exchange football banter with, rather than a concise college professor with all the knowledge of the world, even if it can offer some profound insights and function as pro level research assistant.

In other news from OpenAI, you can now change GPT-5 models manually instead of relying on their automatic picker, rate limits for GPT-5 Thinking is up to 3,000 messages per week, and the context window for it is now 196K tokens — which should be good news for coders.

Read more: Sam Altman’s full announcement on x.com, writeup on TechCrunch.