OpenAI today introduced GPT-5, which the company says brings a "significant leap" in intelligence and fewer hallucinations. GPT-5 is supposed to be better at following instructions and minimizing sycophancy.
The model offers improved performance across coding, math, writing, health, visual perception, and more. OpenAI says that GPT-5 is its best coding model to date, with improvements to complex front-end generation and debugging large code repositories.
Along with better coding, GPT-5 has improved writing capabilities and it is better at answering health-related questions.
According to OpenAI, GPT-5 is "significantly less likely to hallucinate" compared to prior models. With web search enabled, GPT-5's responses are 45 percent less likely to include a factual error than GPT-4o, and when thinking, responses are 80 percent less likely to have errors than OpenAI o3.
GPT-5 is less likely to lie to the user about what it is able to complete, or what has been done. It more honestly communicates its actions and capabilities for tasks that are impossible, underspecified, or missing key tools.
There are four personalities that users can select for GPT-5, including Cynic, Robot, Listener, or Nerd. The different personalities are opt-in and let users set how ChatGPT interacts and responds to questions.
GPT-5 is available to all ChatGPT users, but Plus subscribers will get more usage before a limit is imposed, and Pro subscribers can access GPT-5 Pro, a version that has extended reasoning.
When users hit a usage limit, ChatGPT will switch to a mini version of GPT-5 for remaining queries. GPT-5 replaces GPT-4o, OpenAI o3, OpenAI o4-mini, GPT-4.1, and GPT-4.5.
