The jump from GPT-4 to GPT-5 isn’t nearly as dramatic as previous generations. This pattern is showing up everywhere in AI.
But here’s why I’m excited: We’re moving from “bigger models” to “smarter engineering.”
The companies winning the next phase won’t be those with the biggest compute budgets - they’ll be the ones building sophisticated AI agents with current models.
This shift changes everything about how we should approach AI development.
<a href="https://open.substack.com/pub/diamantai/p/ais-scaling-era-just-ended-whats?r=336pe4&amp;utm_campaign=post&amp;utm_medium=web&amp;showWelcomeOnShare=false" target="_blank" rel="noopener nofollow">read my full analysis</a>

Stay at the forefront of AI with full access to my newsletters on cutting-edge techniques, exclusive tutorials, and extensive archives. Keep up-to-date with our open-source projects: RAG Techniques, GenAI Agents, and Prompt Engineering, and more

DiamantAI

The performance gap between GPT-4 and GPT-5 is smaller than previous generational leaps, signaling the end of the AI scaling era where bigger models automatically meant better performance. The future of AI development is shifting toward sophisticated engineering and AI agents built with existing models, rather than relying on massive compute budgets and larger model architectures.

GPT-5 just proved something important - the scaling era is over

Diminishing returns. Pretty much every AI researcher predicted this.

at the end it’s about prompt engineering and this sucks a lot!

4b to 20b was a a significant jump. Gemini proved that larger models aren’t necessarily going to be better from the economical point of view, due to quality not rising linearly with resources required to run a larger model, but the jump from 4b to 2b “unlocked” some use cases that were not possible at 4b context. This latest jump achieves nearly nothing. In fact I think Open AI knew that, and I think the biggest jump from 4 to 5 is response time speed rather than accuracy or sophistication.

Well, you can only squeeze that much from an analytical computer based on transistors and a binary number system

GPT 5 is so stupid. I tested it on some basic things other than coding. I gave it a customer’s email, demanding a refund due to an issue on the app’s end. Previous models asked for screenshots, proof of the bug etcetera, the current model simply agreed.
GPT 4o sucks too.
I have cancelled my subscription.