What's up, fellow coders? Let's talk about the new shiny toy in the sandbox: GPT-5. The hype has been REAL. We were promised a revolution, a paradigm shift, the AI overlord that would finally understand our spaghetti code & turn it into a Michelin-star-worthy bolognese. & for some, it has been a game-changer. But here's the thing... a quiet but growing chorus of developers, especially those of us living in the super-charged environment of the Cursor IDE, are… well, switching back to an old favorite: Claude Sonnet 4.

It’s a bit of a plot twist, right? The latest & greatest being benched for the tried-and-true. But as with all things in the fast-paced world of AI-assisted coding, the story is a lot more nuanced than just "new versus old." It's about workflow, it's about integration, & honestly, it's about that "vibe" you get when you're in the zone.

So, grab your coffee, pull up a chair, & let's get into the nitty-gritty of why some of the most forward-thinking developers are making this seemingly backward move.

When GPT-5 dropped in Cursor, it felt like Christmas morning. Here was this powerhouse of a model, promising to untangle our most complex logical knots & architect solutions with the wisdom of a seasoned principal engineer. & to be fair, in many ways, it delivers. I’ve seen it solve in minutes a bug that had two senior devs & Claude stumped for a month. The sheer problem-solving capability is, at times, mind-blowing. It can look at a massive monorepo & spot architectural issues & race conditions that other models, including Sonnet 4, might have introduced.

But then… the honeymoon ended. For a lot of us, the day-to-day experience of working with GPT-5 in Cursor started to show some cracks. The most common grumble? It’s… slow. Painfully slow. Like, “go-make-a-sandwich-and-come-back” slow. For simple tasks, it can feel like you’ve brought a tank to a knife fight. There are a lot of complaints about it getting stuck in debug loops, running itself into the ground without ever making progress. One developer even mentioned that it tried to re-run tests over & over without changing a single line of code. That’s not just inefficient; it’s frustrating.

Then there's the "rogue agent" behavior. We've all seen it. You ask it to refactor a small component, & it decides to rewrite your entire runtime from bun to node.js without so much as a "by your leave." Or it will create a whole new design for a landing page when you just asked for a minor tweak. That kind of unpredictability can be a major workflow killer. You spend more time course-correcting the AI than you would have spent just writing the code yourself.

Here's the thing, it's not just about raw intelligence. It's about being a good dance partner. & right now, for a lot of developers in Cursor, GPT-5 feels like it’s stepping on their toes.

So, why are so many developers flocking back to Claude Sonnet 4? It really boils down to a few key things:

This is a big one. Sonnet 4's integration into Cursor feels… right. It's smooth, it's intuitive, & it just works. The "TO DO" list feature, where it breaks down a complex task into a checklist, is a fan favorite. It gives you a roadmap, a sense of structure, & it's something that developers have noted is sorely lacking in the GPT-5 integration. Sonnet 4 seems to understand the context of the IDE better, it uses the tools more effectively, & it doesn't go off the rails as often. It’s like the difference between a bespoke suit & one you bought off the rack. Both might be made of fine material, but one just fits you perfectly.

While GPT-5 is thinking deep thoughts, Sonnet 4 is already on its third or fourth task. For the bread-and-butter coding that makes up most of our days – writing functions, fixing minor bugs, doing small refactors – Sonnet 4 is a speed demon. It’s not just about the raw output speed; it's about the entire interaction. It's less "thinking," more "doing." & in a world of tight deadlines & agile sprints, that matters. A LOT.

This is a fascinating observation I’ve seen pop up in a few forums. Sonnet 4 is often described as a really good junior-to-mid-level developer. It's not going to architect your entire system from scratch (though it can do a surprisingly good job at UI overhauls), but it's a reliable pair programmer. It follows instructions, it doesn't get too creative, & it produces clean, predictable code. GPT-5, on the other hand, is sometimes described as that brilliant but erratic senior dev who might either solve your problem in a stroke of genius or go on a tangent that creates three new ones.

This is an interesting one. While GPT-5 is often praised for its logical prowess, Sonnet 4 seems to have a better eye for design. When it comes to front-end development, Sonnet 4 often produces more aesthetically pleasing results. It’s a subtle thing, but for developers who are also responsible for the look & feel of an application, it’s a significant advantage.

We can't have this conversation without talking about money. & here, things get a little more complicated. On paper, GPT-5 is often cheaper per token than Claude Sonnet 4. For businesses & individual developers watching their budgets, this is a HUGE plus. But here’s the catch: GPT-5's tendency to be overly verbose & to use a lot of tokens in its "thinking" process can sometimes negate that cost advantage. If a model takes three times as many tokens to get to the same solution, is it really cheaper?

This is where the model-agnostic nature of an IDE like Cursor is so powerful. It allows developers to switch between models based on the task at hand. Need to solve a complex architectural problem & don't mind the cost? Spin up GPT-5. Need to crank out some boilerplate code or do a quick refactor? Switch back to the more cost-effective & faster Sonnet 4. It’s about having the right tool for the job, & not being locked into a single ecosystem.

This is something we're seeing across the board in the world of AI. Businesses are realizing that a one-size-fits-all approach doesn't work. That's why platforms like Arsturn are becoming so popular. They allow businesses to build their own no-code AI chatbots, trained on their own data. This means they can create a customer service experience that is perfectly tailored to their brand & their customers' needs, rather than relying on a generic, off-the-shelf solution. It's the same principle as choosing the right AI model in your IDE: customization & control are key.

We can't talk about this without giving a massive shout-out to Cursor itself. This isn't just a text editor with an AI bolted on. It's an AI-first coding environment, & its features are at the heart of this debate. The ability to use symbols to bring in context from files, folders, & even the web is a game-changer. The seamless integration with Git, the ability to run terminal commands directly from the AI chat… these are the things that make modern AI-assisted development possible.

The fact that Cursor allows for this kind of "model-switching" is a testament to its forward-thinking design. They understand that the AI landscape is constantly changing, & that what's best today might not be best tomorrow. By providing a flexible, model-agnostic platform, they're empowering developers to make the best choices for their own workflows.

It's a similar philosophy to how modern businesses are approaching customer engagement. They're not just relying on a single channel anymore. They're using a combination of live chat, email, & increasingly, AI-powered chatbots to provide instant, 24/7 support. A platform like Arsturn is perfect for this, as it allows businesses to create custom AI chatbots that can answer questions, generate leads, & engage with website visitors around the clock, freeing up human agents to handle more complex issues. It's all about using the right tool for the right interaction.

At the end of the day, a lot of this comes down to personal preference & that intangible "vibe." Some developers love the challenge of taming a powerful but wild model like GPT-5. They enjoy the deep, philosophical discussions about code architecture & are willing to put up with the occasional quirkiness to get to that "aha!" moment of brilliance.

Others prefer the steady, reliable companionship of Sonnet 4. They want a tool that gets out of their way & lets them focus on what they do best: building things. They value speed, efficiency, & predictability over raw, untamed power.

There's no right or wrong answer here. It's a spectrum, & where you fall on it depends on your personality, your coding style, & the specific demands of your project.

Honestly, the "GPT-5 vs. Claude Sonnet 4" debate is less about which model is "better" in a vacuum & more about which model is better for you, right now, in your specific workflow.

If you're tackling a really thorny logical problem, or you need to understand the deep architectural underpinnings of a legacy system, GPT-5 is probably your go-to. Its reasoning capabilities are, at times, in a class of their own.

But for the day-in, day-out grind of feature development, UI creation, & general coding tasks, there's a strong case to be made for sticking with Claude Sonnet 4. Its speed, reliability, & seamless integration with Cursor make it a formidable workhorse.

The beautiful thing is, you don't have to choose. The future of coding isn't about pledging allegiance to a single AI model. It's about having a toolbox of powerful, specialized tools & knowing which one to pick for the task at hand.

Hope this was helpful! I'm super curious to hear about your own experiences. Have you made the switch? Are you team GPT-5 or team Claude? Let me know what you think. The conversation is just getting started, & it's a pretty exciting time to be a developer.