OpenAI has unveiled GPT-5, its long-awaited update to its flagship large language model that powers ChatGPT, which chief Sam Altman hailed as a “major upgrade” and a significant step to artificial general intelligence.
The $300bn company on Thursday said the next-generation model was “state of the art” in areas such as mathematics, science and coding. It will be available to all consumers through ChatGPT, with limits for free users.
The release comes almost two-and-a-half years after the company unveiled GPT-4, which delivered a significant leap in capabilities, and investors and users have been eagerly awaiting the San Francisco-based group’s next model.
OpenAI said GPT-5 was particularly proficient at so-called vibe coding, where users can instruct the AI using text prompts to create novel software. “This idea of software on demand will be a defining part of the GPT-5 era,” Altman added.
In the almost three years since ChatGPT launched and brought generative AI to the mainstream, OpenAI has seen its popularity and valuation surge.
The product now has more than 700mn weekly active users and the company is discussing a new valuation of $500bn, according to people with knowledge of the start-up’s plans, which would make it the world’s most valuable private technology company.
OpenAI’s backers have been anticipating a world-beating new model to support that valuation, and justify their faith that the AI group will one day be a multi-trillion-dollar company.
Reaction, however, has been mixed.
“GPT-5 meets expectations in technical performance, exceeds in task reasoning and coding, and underwhelms in [other areas],” said Gartner analyst Chirag Dekate, highlighting that it “does not appear to cross the AGI threshold . . . and early observations seem to indicate incremental progress in writing quality.”
Yet even incremental improvements are significant in OpenAI’s commercial battle with rivals.
Coding has emerged as one of the most widespread and commercially valuable use cases of AI. OpenAI rival Anthropic has dominated the market to date, but GPT-5 marginally outscored Anthropic’s latest model on SWE-bench Verified, an industry-wide test for coding capabilities.
That has won OpenAI important users, including the team behind $10bn start-up Anysphere, whose popular code assistant Cursor operates on top of leading AI models. Michael Truell, the company’s chief executive, described GPT-5 as “remarkably intelligent”.
Cursor is one of Anthropic’s biggest customers, and if its users also migrate to GPT-5 it would provide a meaningful boost to OpenAI’s annual recurring revenue, a metric measuring income from subscriptions.
ARR has already reached $12bn and it forecast to hit $20bn by the end of 2025, according to a person with knowledge of its finances.
Grok 4 Heavy, however, the model created by Elon Musk’s xAI and launched last month, still outperforms GPT-5 on some tests assessing reasoning and knowledge capabilities.
In tests, GPT-5 was more accurate than other Open AI models and less likely to “hallucinate”, or refuse to answer queries. It is also the first flagship model that has been designated as “high” risk for being used to create biological weapons or novel viruses.
“While we do not have definitive evidence that this model could meaningfully help novices to create severe biological harm — our defined threshold for high capability — we have chosen to take a precautionary approach,” the company said.
OpenAI primarily makes money from subscriptions, which range from $20 to $200 monthly for individual users, with custom pricing for businesses or educational institutions.
This week, OpenAI also released its first “open” AI model since it launched ChatGPT, as it attempts to quash rising competition from Chinese start-up DeepSeek and others in the cutting-edge technology.
“The vibes of this model are really good,” said Nick Turley, OpenAI’s head of product for ChatGPT, “But at the end of the day, I think people are just going to have to feel it.”
