GPT – 5 recalibrated: OpenAi admits that “the tone was too cold”, but avoids the real problem

It is not a revolution, but rather a retouching. In a message published this weekend on X, Openai announces that it has changed the personality of GPT – 5, Surveyer modelto return it “West and friendly”. No spectacular emotional turning point, but a series of micro-adjustments supposed to make it “More affordable” In the words of the company.

OpenAi softens the tone of GPT – 5 to meet criticism

The initiative follows a wave of user feedback pointing too formal, sometimes perceived as cold, even distant. Clearly: GPT – 5 was doing the job, but without really making you want to stay in conversation.

The changes are minimal, voluntarily discreet. Formulas like “good question” or “well seen”, appear in certain answers. They do not aim to flatter or seduce, specifies Openai, but to reintroduce a little naturalness, where the model, for the sake of neutrality or efficiency, had ended up erasing all traces of voice.

This correction, even modest, says a lot. She acts that performance is no longer enough. An AI can code, correct, summarize or translate to perfection, if it does not know how to do it with a minimum of tact, it loses its use value. Artificial intelligence is now also played in form. In tone. In the way of being there, even in simulation. And it is not an exaggeration: since the release of GPT-5, the community forums of Openai and Reddit overflow of user messages which go so far as to qualify this transition as “Mentally devastating”.

OPENAI specifies having conducted internal tests: no excess of flattery, as GPT 4O sometimes did itwould not have been detected. No emotional overflow, therefore. Just a way of avoiding what GPT – 5 has embodied a little too well in recent times: clinical perfection, without nuance or warmth.

Cosmetic correction, not structural

The update is being deployed. Invisible, perhaps. Essential, surely. But this façade update, as legitimate as it is, raises another question: why does OpenAi still say nothing on the bottom? On slowness, blockages, more frequent factual errors? Since its launch, GPT – 5 has not only convinced for tone reasons. Many users evoke an unstable model, less reliable than expected, and sometimes behind compared to GPT – 4O.

Nothing, in the official announcement, approach these points. Not a word on performance. As if the real problem, the one who undermines confidence, could be resolved with a well -placed politeness formula. The warm varnish of a wobbly model?

Want to save even more? Discover Our promo codes Selected for you.