So cursor's free gpt-5 week ended today so I figured now's the perfect time to write down my thoughts after using it daily for a week.

The first prompt

When I heard Cursor was offering gpt-5 for free during launch week I immediately jumped into the editor and started building. At the time I was working on a chunking strategy for markdown notes in a rust codebase and this is what my first prompt was:

gpt-5 had me from the get go. I could immediately see and feel the difference with it compared to older OpenAI models and even Claude's sonnet-4. I really liked the solution it implemented and even more the concise but remarkably clear summary of changes it gave me with the next set of improvements to make.

The honeymoon phase

For the first two days gpt-5 was just in the zone implementing things perfectly, suggesting changes and letting me know of edge cases that I did not even think of. It was just a joy to work with.

In this period I never once thought "maybe I should switch models" or "maybe this task is better suited for Sonnet or Opus".

Two important things here:

• I was just using regular gpt-5-thinking at this point. I did not know that max mode (uses the model's full context window + thinks longer) was also free to use.

• I had no idea what the api pricing for the gpt-5 series models were.

I assumed gpt-5 would be about 1.5-2x the price of gpt-4.1 and I'd have been perfectly fine with that. What I wasn’t prepared for was it being lower than gpt-4.1 and the input tokens costing almost 40% less.

I just thought to myself: I might never have to touch Sonnet again.

And honestly that's a relief because with Cursor's new pricing changes I'd burn through my monthly allowance in days with Sonnet.

The rough patch

Around day three or four I switched to max mode on cursor expecting the output quality to be significantly better. But... not much changed. With Sonnet the difference between regular and max is usually instantly obvious. For some reason here it wasn't.

Also as my project got more complex I noticed gpt-5 struggling and sometimes getting stuck in a loop and just generally the responses felt... less intelligent.

Then I saw others on Twitter saying similar things although not everyone had the same experience as me, a lot of people were really liking the quality after switching to max mode.

One thing to note is that during this period cursor pushed an update to fine tune their prompt to "improve the code quality" but that could've made the model performance worse too because a lot of people were specifically saying the model was performing badly on cursor.

Bonus: Blind Test

I took a 20 question blind test where for each question I had to pick between a gpt-4o and a gpt-5 generated answer. And boy was I surprised by how much I preferred gpt-5. The result was 17-3.

TL;DR

GPT-5 made a really great first impression on me and it has an unbeatable price for what it offers (including the mini and nano models) and enough quality to make me reach for it first. But there's surely a lot of things to fix and improve on both Cursor and OpenAI's side and I really hope it stabilizes over time because I'd take gpt-5 on its best day over any other model.