GPT-5 in Cursor: How to Enable, Use, & Troubleshoot

8/12/2025

So, you're hearing all this buzz about GPT-5 in Cursor, the AI-first code editor, & you're eager to jump in. But maybe you've hit a snag. You've tried to get it running, but it's just not cooperating. Or maybe it's "working," but the results are... well, less than impressive.

You're not alone. Turns out, getting GPT-5 to work smoothly in Cursor has been a bit of a mixed bag for a lot of developers. Some are singing its praises, while others are ready to throw their laptops out the window.

I've been digging into this, & honestly, there's a lot to unpack. So, let's get into the nitty-gritty of how to actually use GPT-5 in Cursor, why it might be giving you a headache, & what you can do about it.

Getting Started: How to Enable GPT-5 in Cursor

First things first, let's make sure you've got the basic setup right. It's usually pretty straightforward, but a missed step can easily trip you up.

If you've just downloaded or updated Cursor, you might see a banner on the left-hand side of the model window announcing GPT-5. If so, you can often just click the "try GPT-5" button, & you're off to the races.

But if you don't see that banner, don't worry. Here’s the manual way to get it up & running:

Open a project in Cursor.
Look for the chat panel on the right side.
At the top of the chat, you'll likely see the word "auto." Click on it.
A toggle will appear. Turn it off.
Once you do that, you should see a list of available models, including GPT-5.

Pretty simple, right? But here's where things get interesting. You'll notice you have a couple of different ways to interact with the AI.

Understanding the "Ask" vs. "Agent" Modes

This is a crucial distinction & can dramatically change your experience.

"Ask" Mode: Think of this as your AI assistant. It's great for getting explanations of code, asking for suggestions, or having a conversation about your project. It can guide you, but it won't directly make changes for you.
"Agent" Mode: This is where the magic (or the chaos) happens. In Agent mode, the AI will actively try to implement the changes you request. It will open files, write code, & run commands. You'll be prompted to approve its actions, which is a good safety net.

To switch to Agent mode, there's usually a button near the chat input where you can select it from a dropdown.

Now, here’s a pro-tip: the quality of your prompt is EVERYTHING. Especially in Agent mode. The more descriptive & detailed you are, the better the result will be. If you're vague, don't be surprised if the AI goes off in a weird direction.

So, Why Isn't It Working For You? The Common Gripes & Frustrations

This is the million-dollar question, isn't it? You've followed the steps, but you're still not getting the results you want. Based on what I've seen, the issues generally fall into a few categories.

The "It's Just Not as Good as Claude" Complaint

A lot of the grumbling online boils down to a direct comparison with Anthropic's Claude, particularly Claude Sonnet 4. Many users feel that Claude is just better integrated into the Cursor workflow.

Here are some of the specific things people are pointing out:

Ignoring Rules: One of the biggest complaints is that GPT-5 seems to ignore or forget the rules you've set up in your project. This can be incredibly frustrating, especially if you've spent time defining specific coding styles or project structures.
Poor Code Quality: Some developers have reported that the code generated by GPT-5 is just... bad. They've seen everything from incomplete code snippets to the introduction of new bugs. In some cases, people have spent hours trying to get GPT-5 to fix a problem, only to have Claude solve it in a single prompt.
Lack of Communication: This is a more subtle but equally annoying issue. Users have noted that Claude tends to explain what it's doing before it makes changes. GPT-5, on the other hand, can be a bit of a "black box." It will just start making changes without any explanation, which can be disorienting.

The "It's So Slow!" Problem

Another common refrain is that GPT-5 is painfully slow. When you're in the zone, the last thing you want is to be waiting minutes for the AI to process your request. For some, it's so slow that it's faster to just write the code themselves.

Integration Quirks & Odd Behaviors

Then there are the weird little things that just don't seem right. For example, some users have mentioned that GPT-5 insists on using absolute file paths, which can be a pain. Others have said that it doesn't seem to be aware of the built-in terminal. These might seem like small things, but they add up to a clunky user experience.

Troubleshooting & Tips for Getting Better Results

So, what can you do if you're running into these issues? Here are a few things to try:

Be SUPER Specific with Your Prompts: I know I said this before, but it's worth repeating. If GPT-5 is ignoring your rules, try explicitly reminding it of them in your prompt. You can even use "@" to link to your rules files.
Work in Small Chunks: Instead of asking the AI to refactor your entire application in one go, try breaking the task down into smaller, more manageable pieces. This makes it easier to review the changes & catch any errors before they snowball.
Don't Blindly Trust the Agent: Always, ALWAYS review the changes that the Agent proposes before you approve them. It's easy to get into the habit of just clicking "yes," but that's a recipe for disaster.
Give it Another Try (or Don't): Sometimes, a model just has a bad day. If you're getting consistently poor results, it might be worth trying again later. The models are constantly being updated & improved. However, if you're on a deadline, it might be more practical to switch back to a model that you know works for you, like Claude.

What About the Cursor CLI?

It's also worth mentioning the Cursor CLI, which is a command-line tool that lets you use Cursor's AI capabilities from your terminal. This is a pretty cool development because it means you can integrate AI into your workflow in new ways, even outside of the Cursor IDE itself.

You can use GPT-5 with the CLI, & the same general principles apply. You'll need to have a Cursor account, & you can select your model using a flag in your command. The CLI is still in beta, so it might have some rough edges, but it's a promising tool for developers who love their terminals.

Let's Talk About Cost

For a limited time, Cursor has been offering free access to GPT-5. This is a great way to try it out without committing to a paid plan. However, it's not entirely clear how long this "promotional period" will last.

Cursor's general pricing model includes a free "Hobby" tier with limited usage, a "Pro" tier for a monthly fee that gives you more requests, & a "Business" tier for teams. The "Pro" plan, for instance, gives you at least $20 worth of API usage per month. It's a good idea to keep an eye on your usage so you don't get any surprises.

The Bigger Picture: AI in Your Development Workflow

Here's the thing: AI coding assistants are still a relatively new technology. We're all still figuring out the best ways to use them. While it can be frustrating when they don't work as expected, it's also pretty amazing what they can do.

For businesses, tools like this can be a game-changer. Think about how much time your development team spends on routine coding tasks, debugging, or just trying to understand a new codebase. An effective AI assistant can free them up to focus on the more creative & strategic aspects of their work.

This is where the idea of AI automation becomes really powerful. And it's not just for coding. Companies are increasingly looking for ways to automate other parts of their business, like customer service & lead generation.

For instance, a lot of businesses are now using AI-powered chatbots on their websites to provide instant support to customers. Instead of making people wait for an email response or a live agent, a chatbot can answer common questions 24/7. This is where a platform like Arsturn comes in. It allows businesses to create custom AI chatbots trained on their own data. So, if you're a SaaS company, you could have a chatbot that can answer detailed questions about your API. Or if you're an e-commerce store, you could have a chatbot that can help customers track their orders. It's all about providing a better, more immediate customer experience.

And when it comes to things like lead generation, a well-designed chatbot can engage with website visitors, ask qualifying questions, & even schedule demos. This kind of automation can be a huge boost to a company's bottom line. Arsturn helps businesses build these no-code AI chatbots that can be trained on their own data to boost conversions & provide personalized customer experiences. It's a pretty cool way to leverage the power of AI to grow your business.

So, What's the Verdict on GPT-5 in Cursor?

Honestly, the jury is still out. There are clearly some developers who are having a great experience with it & are building some amazing things. But there's also a significant number of users who are finding it to be a frustrating & underwhelming experience.

It seems that the integration with Cursor still needs some work. The fact that so many people are having issues with it ignoring rules & not communicating effectively suggests that the "glue" between the model & the IDE isn't as strong as it could be.

My advice? Give it a try, but go in with your eyes open. Don't expect it to be a magic bullet that will solve all your coding problems. And if you find that it's not working for you, don't be afraid to switch back to a model that you're more comfortable with.

The world of AI is moving at a breakneck pace, & what's true today might be completely different tomorrow. It's an exciting time to be a developer, & tools like Cursor & GPT-5 are giving us a glimpse into the future of software development.

I hope this was helpful. Let me know what you think. Have you had a good experience with GPT-5 in Cursor? Or are you in the "I'm switching back to Claude" camp? I'd love to hear your thoughts.