Developers Describe GPT-5 as a Mixed Blessing Strong

Some developers report mostly positive experiences with GPT-5 thus far. Jenny Wang, an engineer and investor who created the personal styling assistant Alta, mentioned to WIRED that the model seems particularly adept at tackling complex coding challenges in a single attempt, outperforming previous versions like OpenAI’s o3 and 4o, which she often uses for tasks such as formatting or creating API endpoints similar to existing ones.

During her trials with GPT-5, Wang tasked the model with generating code for a press page for her company’s website, incorporating specific design elements to match the site’s aesthetic. GPT-5 managed to complete the task in one go, a significant improvement over previous experiences where she had to refine her prompts multiple times. However, she did encounter one major issue: “It hallucinated the URLs,” she noted.

Another developer, who preferred to remain anonymous as they hadn’t received permission from their employer to speak publicly, emphasized that GPT-5 shines in addressing profound technical issues.

This developer’s current side project involves creating a programmatic network analysis tool, which necessitates secure code isolation. “I presented my project and some potential approaches, and GPT-5 processed it all, offering several recommendations along with a realistic timeline,” the developer shared. “I’m impressed.”

A few enterprise partners and clients of OpenAI, such as Cursor, Windsurf, and Notion, have openly praised GPT-5’s coding and reasoning capabilities. Notion even mentioned on X that it is “fast, thorough, and handles complex work 15 percent better than other models we’ve tested.”

However, shortly after GPT-5’s launch, some developers took to online platforms to voice their concerns. Many felt that GPT-5’s coding abilities lagged behind expectations for a cutting-edge model from a leading AI company.

“OpenAI’s GPT-5 is impressive, but it feels like technology that could have come out a year ago,” remarked Kieran Klassen, a developer currently working on an AI email assistant. “Its coding skills remind me of Sonnet 3.5,” he added, referencing an Anthropic model released in June 2024.

Amir Salihefendić, founder of Doist, expressed in a social media post that his experience with GPT-5 in Cursor has been “pretty underwhelming,” particularly regarding its coding abilities. He likened the launch of GPT-4 to a “Llama 4 moment,” referring to Meta’s AI model, which also disappointed some in the AI community.

Developer Mckay Wrigley commented on X that while GPT-5 excels as a “phenomenal everyday chat model,” he would continue using Claude Code + Opus for coding tasks.

Other developers have described GPT-5 as “exhaustive”—occasionally helpful but often frustratingly verbose. Wang, while satisfied with the frontend coding project she assigned GPT-5, noted that it was “more redundant” than necessary: “It could have produced a cleaner or shorter solution.” Kapoor pointed out that users can adjust GPT-5’s verbosity, allowing it to be less chatty or to minimize reasoning for better performance or affordability.