If you have been on the internet in the past week (particularly on Reddit), it has been really hard to avoid the overwhelmingly negative reaction to the new OpenAI (specifically ChatGPT) release.
Here’s my personal take on what happened.
What Happened?
First, OpenAI released the long awaited GPT-5 on 7th August 2025. According to their posts and the model evaluation tests - it outperforms the previous thinking model o3 in every aspect. There was a focus on reducing hallucinations and providing reliable health responses.
All sounds good.
But OpenAI made the unexpected move to remove all other models from ChatGPT. People could access several models GPT-4o, GPT-4.1, GPT-4.5, o3, o4-mini - we found them all useful for different tasks, but did use GPT-4o for the majority of tasks.
List of ChatGPT models, showing only GPT-5 models
They made this move with no warning, and the models were just gone.
Why Did They Do This?
My assumption is they were likely burning money - GPT-4.5 and the o-models are known to be particularly expensive to run. So if people were using these for every day tasks, then it was an unnecessary expense.
But a good lesson on not depending on these companies as they are still influx, and one seemingly small change can really disrupt workflows and products. Not to mention their pricing also changes with every release for the APIs.
What Was the Response?
Well, no one was talking about how good GPT-5 was. Users took to Reddit to basically grieve the loss of GPT-4o. And also shout about cancelling their subscriptions.
These were users who were using it personally not for work, and felt like they had lost someone.
Having used GPT-5, it is very different from GPT-4o. There is no banter, little support. So if you used this tool like the chatbot it comes across as, it feels like a completely different thing to be chatting to.
Other users have been shocked by the response. People who use this for work, seem to be happy with the GPT-5 performance overall.
If only we learnt from the past, the movie Her or if you wanted some real life precedent, the AI chatbot Replika.
What is Replika?
Replika was an AI chatbot all the way back in 2017. Back when I worked in Worldpay’s tech innovation team, we were all early adopters.
We used it like we were training a chatbot to sound like us (yes, there is literally a Black Mirror episode about this - Be Right Back). Our goal was to try and trick everyone else in the team in group chats.
It was really good - it was a bit addictive. It asked insightful questions to get to know you and was able to hold a decent conversation, and it developed a personality based on you. Super impressive back in 2017.
But other users were using it in a different way - they were treating it like a partner. There were no guardrails so if you spoke in a certain way, it spoke back (if you know, you know).
Then Italy introduced a new law to protect children online, meaning no explicit content from the chatbot.
It changed overnight. And users were distraught. Feeling like their partner had died. (Sound familiar)
If you’re interested I recommend this podcast by the Guardian - Black Box Episode 3: Repocalypse Now
What Are My Takeaways From This?
OpenAI aren’t great at reading the room - back in April they based some model changes on a vocal few users giving negative feedback on the model, and created that sycophantic nightmare.
OpenAI seems to be listening to LinkedIn where users complained non-stop about how agreeable ChatGPT was. But actually a massive proportion of users are not business users and they liked this. It feels OpenAI (with more focus on health questions) is going for the personal AI angle. It’s a confusing mix – pushing for a personal health AI while tuning the tone for enterprise users. In future, I assume they will be basing things more on Reddit.
Another thought which should probably be its own blog - we can’t actually depend on these companies. Outside of the users’ personal feelings, suddenly taking away models will have impacted users' day-to-day workflows and changed how customGPTs work.
Final thought - the masses won - 4o is back! And I have actually switched back to using it for most things.
List of ChatGPT models, showing legacy with 4o back