Builders Say GPT-5 Is a Blended Bag

Some builders say they’ve had largely optimistic experiences with GPT-5 to date. Jenny Wang, an engineer, investor, and creator of the private styling agent Alta, instructed WIRED the mannequin seems to be higher at finishing advanced coding duties in a single shot than different fashions. She in contrast it to OpenAI’s o3 and 4o, which she makes use of incessantly for code era and simple fixes “like formatting, or if I wish to create an API endpoint much like what I have already got,” Wang says.

In her assessments of GPT-5, Wang says she requested the mannequin to generate code for a press web page for her firm’s web site, together with particular design parts that may match the remainder of the location’s aesthetic. GPT-5 accomplished the duty in a single take, whereas previously, Wang would have needed to revise her prompts throughout the course of. There was one important error, although: “It hallucinated the URLs,” Wang says.

One other developer, who spoke on the situation of anonymity as a result of their employer didn’t authorize them to talk to the press, says GPT-5 excels at fixing deep technical issues.

The developer’s present interest venture is writing a programmatic community evaluation device, one that may require code isolation for safety functions. “I principally introduced my venture and a few paths I used to be contemplating, and GPT-5 took all of it in and gave again a number of suggestions together with a sensible timeline,” the developer explains. “I’m impressed.”

A handful of OpenAI’s enterprise companions and prospects, together with Cursor, Windsurf, and Notion, have publicly vouched for GPT-5’s coding and reasoning abilities. (OpenAI included many of those remarks in its personal weblog publish saying the brand new mannequin). Notion additionally shared on X that it’s “quick, thorough, and handles advanced work 15 p.c higher than different fashions we’ve examined.”

However inside days of GPT-5’s launch, some builders have been weighing in on-line with complaints. Many mentioned that GPT-5’s coding talents appeared behind-the-curve for what was speculated to be a state-of-the-art, ultra-capable mannequin from the world’s buzziest AI firm.

“OpenAI’s GPT-5 is excellent, nevertheless it looks as if one thing that may have been launched a 12 months in the past,” says Kieran Klassen, a developer who has been constructing an AI assistant for e mail inboxes. “Its coding capabilities remind me of Sonnet 3.5,” he provides, referring to an Anthropic mannequin that launched in June 2024.

Amir Salihefendić, founding father of the startup firm Doist, mentioned in a social media publish that he’s been utilizing GPT-5 in Cursor and has discovered it “fairly underwhelming” and that “it’s particularly unhealthy at coding.” He mentioned the discharge of GPT-4 felt like a “Llama 4 second,” referring to Meta’s AI mannequin, which had additionally disenchanted some individuals within the AI group.

On X, developer Mckay Wrigley wrote that GPT-5 is a “phenomenal on a regular basis chat mannequin,” however relating to coding, “I’ll nonetheless be utilizing Claude Code + Opus.”

Different builders describe GPT-5 as “exhaustive”—at instances useful, however usually irritating in its long-windedness. Wang, who total was happy total with the frontend coding venture she assigned to GPT-5, says that she did discover the mannequin was “extra redundant. It clearly might have give you a cleaner or shorter resolution.” (Kapoor factors out that the verbosity of GPT-5 may be adjusted, in order that customers can ask it to be much less chatty and even do much less reasoning in alternate for higher efficiency or cheaper pricing.)