Some programmers say that so far they have largely had positive experience with GPT-5. Jenny Wang, engineer, investor and creator of the Alta stylization agent, said Wired that the model seems to be better in performing sophisticated coding tasks in one shot than other models. She compared it to O3 and 4o OpenAI, which she often uses to generate code and basic corrections “like formatting or if I want to create an API end point similar to what I already have,” says Wang.
In her GPT-5 tests, Wang said that she asked the model to generate a press page code for its company’s website, including specific design elements that would match the rest of the site. GPT-5 performed the task for one shot, while in the past Wang would have to revise his hints during this process. However, there was one significant mistake: “he hallucinated URL addresses,” says Wang.
Another developer who spoke provided anonymity, because their employer did not authorize them to talk to the press, says GPT-5 is distinguished by solving deep technical problems.
The current design of the programmers hobby is to write a program tool for network analysis that would require code insulation for safety purposes. “I basically presented my design and some paths that I considered, and GPT-5 took it all and gave several recommendations along with a realistic timeline,” explains the developer. “I am impressed.”
A handful of OPENENAI Enterprise partners and clients, including cursor, windsurf and concept, publicly managed the skills of GPT-5 coding and reasoning. (OpenAI contained many of these comments on its own Blog post Announcement of a modern model). Concept too Shared at X That it is “fast, accurate and supports complex work 15 percent better than other models that we tested.”
But within a few days of the GPT-5 release, some programmers weighed online with complaints. Many said that the GPT-5 coding abilities seemed to be buried because of what was supposed to be the most state-of-the-art, ultra, unusual model from the most buzzing AI company in the world.
“GPT-5 OPENAI is very good, but it seems that there is something that was released a year ago,” says Kieran Klassen, a developer who built AI assistant in the e-mail inboxes. “His coding capabilities remind me of Sonnet 3.5”, he adds, referring to the anthropic model, which was launched in June 2024.
Amir Saliihefendić, founder of the doist startup company, – he said in a post on social media that he used the GPT-5 in the cursor and considered it “quite disappointing” and that “he is particularly bad in coding.” He said that the release of the GPT-4 seemed “moment of Lama 4”, referring to the AI Meta model, which also disappointed some people in the AI community.
On X, McKay Wrigley programmer wrote This GPT-5 is a “phenomenal model of chat every day”, but when it comes to coding: “I will continue to use Claude Code + Opus.”
Other programmers define GPT-5 as “exhausting”-in helpful time, but often annoying in its long-lasting. Wang, which was generally satisfied with the Frontend coding project, which was assigned to GPT-5, claims that she noticed that the model was “more unnecessary. Apparently he could come up with a cleaner or shorter solution.” (Kapoor emphasizes that GPT-5 has been adjusted so that users can ask them to be less talked or even less reasoning in exchange for better performance or cheaper prices.)