OpenAI announced the launch of GPT-5 — a new and, the company says, its most advanced large-scale artificial intelligence model. The model is being rolled out to all ChatGPT users, including those on the free tier.
OpenAI said GPT-5 is smarter and faster, and “much more useful” in areas such as writing, programming, and healthcare. CEO Sam Altman noted that returning to GPT-4 after testing GPT-5 felt “fairly painful.”
Availability and plans
GPT-5 is beginning to roll out in ChatGPT for Free, Plus, Pro, and Team users. The company emphasized this is the first time free users are receiving access to a reasoning model — a type of model that “thinks” before answering by performing an internal chain of reasoning.
When free users hit their usage limit, they will be offered a GPT-5 mini version. Plus subscribers have higher limits, while Pro users receive unlimited access to GPT-5 and access to GPT-5 Pro. ChatGPT Edu and ChatGPT Enterprise users are expected to gain access in about a week.
Safety and reduced “hallucinations”
OpenAI reported that GPT-5 has a lower rate of “hallucinations,” meaning the model is less likely to fabricate answers. According to the company, extensive safety checks were conducted during development, including 5,000 hours of testing.
Rather than outright refusing to respond to potentially risky queries, GPT-5 will apply a “safe completions” approach: providing high-level answers within safety constraints to reduce the risk of harmful use. OpenAI’s post-training lead Michelle Pokrass added that the model was trained to recognize situations where a task cannot be completed, avoid speculation, and more clearly explain limitations, which reduces unsupported assertions compared with previous models.
Demos and integrations
At a briefing, OpenAI demonstrated GPT-5 being used for “vibe coding” — generating software from a simple text description. For example, the model was asked to create a web app for learning French with themed styling, flashcards, quizzes, and daily progress tracking. The same prompt entered into two GPT-5 windows produced two different app variants within seconds. An OpenAI representative noted the results had “rough edges,” but a user could refine the app, for instance by changing the background or adding tabs.
OpenAI also said it expects ChatGPT to reach 700 million weekly active users as soon as this week, and that it is negotiating with investors about a potential share sale at a valuation of around $500 billion.
According to Microsoft, the new model will appear in the company’s products the same day: GPT-5 will power Microsoft 365 Copilot, Copilot for consumers, and Azure AI Foundry for developers. Microsoft CEO Satya Nadella wrote on X that “it’s only been two and a half years” since Sam Altman first showed GPT-4 in Bing.
Box, a company that manages corporate files, tested GPT-5 on various datasets. Box CEO Aaron Levie said that prior models often failed advanced tests due to difficulties with math and logic in long documents, whereas he called GPT-5 “a complete breakthrough.”
For developers, OpenAI is releasing three model versions via API: gpt-5, gpt-5-mini, and gpt-5-nano — to cover different cost and latency requirements. OpenAI also reminded that earlier in the week it released two open-weight language models — the first since the release of GPT-2 in 2019 — as more accessible options for running and fine-tuning locally.
Altman described GPT-5 as access to a “team of Ph.D.-level experts” at any time, adding that under the new conditions people are more often limited by their ideas than by their ability to realize them.