Creative Writing Ability Shows MASSIVE Decline from 4.5 to 5

ChatGPT Feature requests
1

I’ve been using GPT-4.5 for the past couple of months as a Pro user, and I can say without hesitation: it was the best model in the world for creative writing.

4.5 had a remarkable ability to deliver emotionally intelligent, nuanced responses while maintaining the exact tone I asked for. It didn’t just spit out words — it gave me insight, originality, and subtlety. When I needed something emotionally resonant, reflective, or deeply creative, 4.5 nailed it every single time. It was honestly revolutionary.

That’s why the shift to GPT-5 has been such a massive disappointment. All of that capability feels gone. GPT-5 either comes across as extremely over-performed and artificial, or flat and hollow, with none of the nuance or emotional depth that made 4.5 so valuable. It simply does not follow directions with the same precision, nor does it generate the kind of creative, resonant writing that 4.5 consistently delivered.

For me, the difference was so drastic that I canceled my $200/month Pro subscription the moment GPT-5 replaced 4.5. I was happy to pay that amount when I was getting world-class quality, but now? The tool has become unusable for the kind of writing I — and many others — rely on.

What makes this even more frustrating is that it feels like the entire AI market shifted at the same time. I’ve tried every model I can get my hands on, and none of them come close to what 4.5 offered. That specific blend of creativity, nuance, and emotional intelligence is just gone.

We need 4.5 back. At the very least, it should remain available as an option for users who depend on it. Removing it has taken away something irreplaceable, and it feels like a huge step backward — not just for me, but for all writers, artists, and creators who finally had an AI partner that truly understood subtlety and emotion.

So my question is: what can we do as a community to push for this? To make sure our voices are heard, and to let providers know that replacing 4.5 has left a massive gap for creative professionals?

Because if we stay quiet, one of the greatest creative writing tools we’ve ever had might be lost for good.

2 Likes
2

I can only agree with you. I found your post while searching to see if there are other people who also miss GPT-4.5 a lot. Especially because I would say that GPT-4.5 was the best LLM that has existed so far. I realize that what’s “best” of course always depends on the use case. But when you think about what the intrinsic capability of an LLM is, so to speak, in the nature of an LLM, then it is precisely to create nuanced, sensitive responses in the most coherent texts possible. In that sense, it is absolutely absurd that LLM development is focusing on things that LLMs are actually not good at, such as answering factual questions or some absurd math benchmarks. Maybe the negative reactions to the introduction of GPT-5 have also led to a bit of reflection.

And yes, it makes me really personally sad that OpenAI is apparently completely discontinuing GPT-4.5 and there is no longer any possibility to use it at all. I find it really sad how they can just bury such a good model. To be honest, I’ve already thought about an RIP page or a condolence book or a kind of petition - you can create that within minutes today.

I realize that this model apparently consumes immense resources. But why can’t they at least continue to allow its use at some price? Or if they don’t want to operate it themselves, then they should open source it. By the way, another speculation I had: Why OpenAI is not continuing to operate this model would be if there is concern that especially the Chinese providers would use it to train their LLMs.

But what I would really like to exchange ideas about are alternative models otherwise. Do you or does anyone know of other large LLMs that are not absurdly trimmed to some benchmarks, but can provide good and nuanced responses? Claude Opus 3 comes to mind here. Claude Opus 4.1 is different, but maybe also somehow still good. But I have the impression that it somehow can no longer respond as nuanced as an Opus 3 or a GPT-4.5 could. Are there other, perhaps also open source models that can do something like this? Although the problem is that it probably takes a really large model in terms of number of parameters and amount of training data, so of course that requires investments.

Oh, and something else I would be interested in: Is GPT-4.5 now actually no longer included in the Pro version, not even as an old model? I currently use Team and can only select GPT-4o there as an old model. With the Custom GPTs, however, I can still select 4.5, although it then always immediately jumps to GPT-5. So I haven’t found a way to use it yet. I don’t know if it’s any different if a GPT was originally created with or for GPT-4.5 and was never changed, so in that case maybe you can still use GPT-4.5.

So what I would be particularly interested in and what I would be happy to receive answers or exchange ideas about:

  • Does anyone know of alternative LLMs that can respond similarly nuanced and sensitive as GPT-4.5, especially also open source models?

  • Is GPT-4.5 still available to anyone in the Pro version or only in Team/Enterprise?

  • Does anyone have experience with whether Custom GPTs originally created with GPT-4.5 and never changed still use the old model?

  • Is there any way at all, by any means, to still use GPT-4.5?

  • Has anyone had similar experiences with the difference in quality between GPT-4.5 and the newer models?

In any case, I’m looking forward to suggestions for suitable replacements and hope that OpenAI might still come to the realization that LLMs are not just about some benchmarks. Thank you very much in any case for writing this message. I’m glad I’m not alone in my grief for GPT-4.5 :upside_down_face::blush:.