OpenAI releases GPT-5: Fewer errors, more performance

OpenAI has introduced GPT-5 – a new AI model that, according to the company, marks a significant advance in intelligence, reliability, and versatility. The model is now available in ChatGPT and brings numerous improvements, including error rate, typing quality, health question handling, and software development.

Since the release of GPT-3 in 2020, OpenAI has pushed the boundaries of performance with each new model generation. With GPT-5, the company aims to continue this development consistently. The focus is not only on technical advancements but also on clearer communication of model boundaries and new options for dialog style.

Significantly fewer hallucinations

A central goal of GPT-5 is to significantly reduce the so-called hallucination rate—that is, the number of false or fabricated statements. According to OpenAI, GPT-5 answers with web search enabled contain 45 percent fewer factual errors than those of GPT-4o. In complex reasoning tasks, the error rate is even reduced by 80 percent compared to the o3 model. The AI is also more transparent about its limitations and is more likely to avoid making false promises about its capabilities.

Better performance in many areas

According to OpenAI, GPT-5 improves in several key areas, including:

Coding: Software development benefits from better front-end generation and precise error analysis, even in large code repositories. According to OpenAI, GPT-5 is considered the company's best coding model to date.
Writing: The text quality has been noticeably improved, both stylistically and structurally.
Mathematics: Calculations and complex mathematical operations are solved more reliably.
Health: GPT-5 can answer medical questions in a more differentiated and well-founded manner.
Visual perception: In combination with image data, the model analyzes content more precisely than its predecessors.

These advances are intended to expand the possible applications in research, everyday life and work and make them more reliable.

Customizable personalities

A new feature allows you to choose between four predefined personalities that influence your response behavior: Cynic, Robot, Listener, or Nerd. These variants change your tone and communication style—but the core messages remain the same. Personality selection is optional and is aimed at users who value certain expressions or conversational styles.

Accessibility and use

GPT-5 is now available to all ChatGPT users. Plus subscribers can use the model for a longer period without restrictions. Pro subscribers get access to GPT-5 Pro, an enhanced version with greater reasoning power. When a usage limit is reached, ChatGPT automatically switches to the mini version of GPT-5. With the introduction of GPT-5, OpenAI is replacing several previous models: GPT-4o, OpenAI o3, OpenAI o4-mini, GPT-4.1, and GPT-4.5 are no longer actively used.

GPT-5 brings progress into everyday life

GPT-5 marks a further step in the development of powerful, versatile, and transparent AI systems. OpenAI places a clear focus on content quality, reliability, and user control. The improvements in hallucinations, the new personality options, and the stronger performance in coding, math, and health demonstrate that GPT-5 is not just a technical upgrade, but a real tool for everyday life. (Image: Shutterstock / Diego Thomazini)