The developers say GPT-5 is a mixed bag

Some developers say they have had a largely positive experiences with GPT-5 so far. Jenny Wang, an engineer, investor and creator of the Uta personal design agent, told Alta that WIRED seems to be the model better in completing complex coding tasks in one shot of other models. I compared it to Openai’s O3 and 4O, which it uses frequently to generate code and direct repairs “such as coordination, or if I want to create an API’s end point similar to what I already have,” says Wang.

In her GPT-5 tests, Wang says she asked the model to create a symbol for a press page for her company’s website, including specific design elements that would match the rest of the site’s aesthetic. GPT-5 has completed the task in one, while in the past, Wang had to revise her claims during the operation. There was one important mistake, though: “URL has been stuck,” says Wang.

GPT-5 says another developer, who spoke on the condition of anonymity because the employer was not allowed to speak to the press.

The current Hobby project of the developer writes a software analysis tool, which requires the code isolated for safety purposes. “I mainly made my project and some of the paths I was thinking about, and GPT-5 took everything and returned some recommendations alongside a realistic schedule,” the developer explains. “I admire.”

A handful of institutions and institutions in Openai, including the indicator, Windsurf, and the concept, have publicly performed coding and thinking skills in GPT-5. (Openai included many of these notes on its own Blog post Announcing the new model). Also idea Common to x “It is fast and comprehensive and takes 15 percent better complicated by the other models we tested.”

But within days of the launch of the GPT-5, some developers weighed online with complaints. Many said that the coding capabilities in the GPT-5 looked behind the curve because it was supposed to be a very capable modern model of the world’s artificial intelligence company.

“The GPT-5 of Openai is very good, but it seems that something that could have been released a year ago,” says Kiran Classen, the developer who built an Amnesty International Assistant for Email. “Its coding capabilities remind me of Sonnet 3.5,” adding, referring to the Anthropor model launched in June 2024.

Prince of Scanvic, founder of the Dost starting company, He said in a social media post He used GPT-5 in the indicator and found it “disappointing” and this is “especially bad in coding.” He said that the launch of the GPT-4 felt as if it were “Llama 4”, referring to the Meta model of artificial intelligence, which was disappointed by some people in the artificial intelligence community.

On x, developer Mckay Wrigley books The GPT-5 is “a huge daily chat model”, but when it comes to coding, “I will continue to use Claude Code + OPUS.”

Other developers describe GPT-5 as “comprehensive”-at useful times, but often annoying in its long time. Wang, who was generally pleased with the project to codish the front end she appointed in GPT-5, said she noticed that the model was “more extravagant. It was clear that he could have reached a cleaner or shorter solution.” (Kapoor notes that the GPT-5 verb can be modified, so that users can ask to be less-to-sized or even do less reasons for better performance or cheaper prices.)