OpenAI launched its latest foundational model, GPT-5, on August 7, 2025. OpenAI CEO Sam Altman described the new AI as “our smartest, fastest, most useful model yet,” equating its capabilities to “a legitimate PhD level expert in anything any area you need on demand.” The model was presented as a unified system designed to automatically route user queries to the most appropriate underlying intelligence, immediately rolling out to ChatGPT users and via OpenAI’s API. However, the release generated significant user dissatisfaction and critical commentary, leading OpenAI to implement rapid adjustments, even as prominent partners like Microsoft and Apple moved to integrate the new technology.

GPT-5 is built around a dynamic routing system, directing user requests to gpt-5-main for swift responses or gpt-5-thinking for more complex reasoning. OpenAI reported improvements in reducing factual inaccuracies, stating gpt-5-thinking makes over five times fewer factual errors than its predecessor and shows a 78% reduction in responses with major factual errors. Beyond performance, features included enhanced coding, improved writing, and new custom personalities. A new verbosity control parameter for developers also offers finer control over response length. OpenAI also highlighted extensive safety enhancements, though the encouragement to trust large language models for complex health recommendations remains a concerning area given the critical nature of medical advice.

Despite OpenAI’s optimistic outlook, the GPT-5 launch encountered immediate user dissatisfaction. Many users, particularly on Reddit and X (formerly Twitter), described GPT-5 as a “downgrade.” X user Andres Franco stated, “GPT 5 has been a huge letdown, way more than I expected.” Users often contrasted GPT-5 with GPT-4o, noting a perceived loss of GPT-4o’s “warmth” and “personal” style, describing GPT-5 as “colder” or “mechanical.” Reports included instances of shorter, less accurate responses, and some users highlighted basic mathematical errors.

The live presentation also drew criticism, with some observers describing the graphs used as “sloppy” or “vibe-charting.” AI critic Gary Marcus assessed the launch directly, stating, “It’s here, finally, but not everything people dreamt it would be.” He concluded, “GPT-5 is obviously not AGI,” reflecting ongoing debate about the term Artificial General Intelligence. User feedback was so pronounced that an unnamed X user commented, “The saddest thing in my day is that @garymarcus is right.”

In response to the feedback, OpenAI CEO Sam Altman acknowledged a “bumpy” launch and admitted they “underestimated how much some of the things that people like in GPT-4o matter to them.” In a swift adjustment, OpenAI announced the reintroduction of GPT-4o as a selectable option for paid users and committed to doubling GPT-5 rate limits. Altman attributed some early performance issues to a “broken autoswitcher” bug, stating that GPT-5 would “seem smarter” moving forward.

Despite the challenging debut, major technology companies proceeded with integration. Microsoft swiftly adopted GPT-5 across its Copilot, Microsoft 365 Copilot, and GitHub Copilot offerings. Apple also confirmed its integration into Apple Intelligence, anticipated in early September with iOS 26 and macOS Tahoe 26, for Siri queries and writing tools. Third-party tools like Cursor also quickly made GPT-5 available.