OpenAI just dropped GPT-4.5, which seems to excel in writing

OpenAI just dropped ChatGPT 4.5, the latest upgrade to its flagship AI assistant.

OpenAI CEO Sam Altman made the announcement on X, adding a cryptic detail:

a heads up: this isn’t a reasoning model and won’t crush benchmarks. it’s a different kind of intelligence and there’s a magic to it i haven’t felt before. really excited for people to try it!

Altman didn't clarify what he meant by this, but after reading OpenAI's official announcement and experimenting with the model myself, I think I get what he's hinting at.

But first, let's cover the basics of the new release.

In a nutshell, here's what's new:

4.5 has twice the context

The update dramatically boosts ChatGPT's context — its "memory" of the ongoing history of each conversation.

The model can now keep track of conversations nearly twice as long as before, thanks to scaling advancements in unsupervised learning.

Faster and more intuitive responses

OpenAI claims GPT-4.5's response speed has improved significantly.

Additionally — and this touches on Altman's cryptic statement about a “different kind of intelligence” — the model reportedly has more understanding of nuance and stronger emotional intelligence (EQ) than its predecessors, which makes conversations feel more natural and intuitive:

Enhanced reliability:

OpenAI reports a significant reduction in hallucinations compared to previous models.

They attribute this to advancements in training techniques that bypass the need for the kinds of “reasoning” functionality used in their o1 and o3 family of models:

Where does GPT-4.5 excel?

To follow up real quick on Altman's statement that there's a “magic” to GPT-4.5, I'll just lend my own suggestion that he's referring to the model's ability to write.

In short, it's really good:

Dawn Analytics founder Ben Hyleck shared a side-by-side comparison of texts written by GPT-4.5 and GPT-4o:

I've got lots of fiction writing experience, and I agree with Hyleck: GPT-4.5 is a night-and-day better writer than GPT-4o. (I also think it's better than Claude 3.7 Sonnet, which I tested against the exact same prompt.)

What critics are saying

A common complaint is about the model's tendency toward "over-refusals," meaning it sometimes declines to answer questions even when it could provide a helpful response.

It's also unexpectedly slow — either because of its compute-intensive design or the fact that OpenAI has only freshly rolled it out to hundreds of thousands of ChatGPT Pro subscribers.

Which leads to the final criticism: OpenAI has only rolled GPT-4.5 out to subscribers of the $200 a month pro plan. Altman addressed this point in his announcement post:

bad news: it is a giant, expensive model. we really wanted to launch it to plus and pro at the same time, but we've been growing a lot and are out of GPUs. we will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i'm pretty sure y'all will use every one we can rack up.)