OpenAI has officially launched its new flagship language model GPT-5, which is already available for ChatGPT users and developers via API, reports Baltimore Chronicle with reference to the American tech website The Verge.
According to OpenAI’s CEO Sam Altman, GPT-5 represents a significant breakthrough compared to previous generations. He compared it to the revolutionary leap of the first iPhone with Retina display. “GPT-3 was like a conversation with a school student — sometimes it gave the right answer, sometimes something strange. GPT-4 is more like a university student. And GPT-5 is the first time when the model feels like a PhD-level expert,” said Altman during a press briefing.
OpenAI claims that GPT-5 is more accurate, faster, and makes fewer mistakes. Despite ChatGPT’s weekly user base approaching 700 million, OpenAI has not had the most advanced model on the market since the release of GPT-4. Now, according to Altman, GPT-5 will restore the company’s leadership, especially in key sectors such as medicine, programming, and writing texts. “This is the best model in the world for writing code, writing texts, and for many other areas,” he noted.
One of the key innovations in GPT-5 is the removal of manual model selection in ChatGPT. Now, the model works as a unified system, automatically switching to a more advanced version when necessary — for example, when the user asks the model to “think carefully.” “The old model selection interface was a real puzzle,” admitted Altman.
GPT-5 is already available to all ChatGPT users. However, free account users are limited in the number of queries, after which the model will automatically switch to a lighter version — GPT-5 mini. For developers, OpenAI offers three versions: GPT-5, GPT-5 mini, and GPT-5 nano, each with different performance levels and pricing. Additionally, new “personalities” for the models will appear in ChatGPT: Cynic, Robot, Listener, and Nerd, while users will also be able to choose custom interface colors for specific chats.
OpenAI believes that GPT-5 marks the beginning of the “on-demand software” era. During the company’s tests, the new model outperformed all competitors on key benchmarks for programmers — SWE-Bench, SWE-Lancer, and Aider Polyglot. Yann Dubuay, head of OpenAI’s post-training division, demonstrated how GPT-5 created a French language learning website with an interactive game in just seconds. The model autonomously wrote hundreds of lines of code, and everything worked correctly during the Zoom demonstration.
Earlier we wrote that ChatGPT conversations may be revealed in court, warns OpenAI CEO.