is it just me or something is off in version 2.4.28
feels like its missing the context and doesnt understand my questions
I did notice a very huge decrease with GPT-5.1 Codex Mini which is very clearly tied to the 2.4.28 update. It changed and broke precisely after the update. I liked this model because it was most of the time doing what I expected for a very low token usage / price.
To give some context, I am working on a Laravel project, with Laravel Boost to enhance agent efficiency. The model suddenly appeared unable to navigate the codebase or make simple edits to files. It would begin launching dumb console commands for anything, and even start writing python and node scripts from nowhere, which makes no sense in the PHP project and for the given tasks. I have started many agent conversations and it displayed this behavior quite consistently over different tasks, which is something it never did before, or at least only in some weird hallucination case, but certainly not on simple fresh tasks. I noticed as well that it was very slow and looped a lot about simple steps.
I switched to GPT-5.2 Codex and the behavior is normal.
I am suspecting that something may be broken with the subagents or with another feature that destroyed the 5.1 Codex Mini stability.
On a side note I have an MCP (laravel-boost mcp) which falls in error for no apparent reason quite often. I have to go in the settings and switch it off and on again to restore it. But this was already the case with the previous version.
Overall I would describe my experience with Cursor upgrades as very bad. I was on ‘Early Access’ mode until some point. The regular upgrades should have been moments of excitement, but it was the opposite. Almost every upgrade broke something that worked, sometimes making me lose quite some time to figure out what happened. There seems to be a constant flow of regressions and instabilities with each version. The constant changes in the design and features of the product is also making the experience feel amateurish. I am now in ‘Default’ mode, but even then, the upgrades are often causes for problems.
I hope to land on a good compromise of a model which does not drain tokens, while being reliable and consistant in its behavior, and having Cursor perform with a good level of consistency and transparency. But I do not hope too much.
I completely agree.There seems to be a constant flow of regressions and instabilities with each version.It’s incredibly frustrating — even as a devoted Cursor user, I’m running out of patience waiting for them to fix this.They always push out updates first, then fix bugs later.
1 Like
No, its not just you.
Somebody messed up something. It suddently went from smart to having remembering issues.
Even things that happened few steps ago.