OpenAI has officially launched GPT‑5, its latest flagship model, making it available to all ChatGPT users and developers. CEO Sam Altman described the release as a “dramatic leap” — comparing it to the shift to a Retina display, something users would never want to go back from.
“We are introducing GPT‑5, our best AI system yet. GPT‑5 is a significant leap in intelligence over all our previous models, featuring state-of-the-art performance across coding, math, writing, health, visual perception, and more. It is a unified system that knows when to respond quickly and when to think longer to provide expert-level responses. GPT‑5 is available to all users, with Plus subscribers getting more usage, and Pro subscribers getting access to GPT‑5 pro, a version with extended reasoning for even more comprehensive and accurate answers,” the company announced in an official statement.
Rather than offering separate models for standard and reasoning tasks, OpenAI has built a smart router into ChatGPT. This system automatically sends complex queries to a reasoning-optimized version, without requiring manual switching—even allowing users to trigger it with a simple “think hard” command. Nick Turley, head of ChatGPT, said that the model’s “vibes” are strong and “average users will feel the difference,” thanks to a more intuitive and seamless experience.
OpenAI is also refining access across tiers. Free users will gain access to GPT‑5 up to an undisclosed limit before OpenAI redirects them to GPT‑5 Mini. If you are a Plus subscriber, then you get some broader access, while Pro users unlock GPT‑5 Pro, an even more powerful variant. Developers can access GPT‑5, Mini, and Nano versions via API at different pricing levels. The update brings new personalization features, including four personality modes — Cynic, Robot, Listener, and Nerd — allowing users to tailor ChatGPT’s tone. Thread-specific color accents further boost the user experience.
GPT‑5 is already setting new standards across coding, logic, and health applications. It outperforms previous models on benchmarks such as SWE‑Bench (scored 74.9%), SWE‑Lancer, and Aider Polyglot (scored 88%), and even demonstrates excellence in health-specific evaluations like HealthBench (~ 46.2%). At the launch, GPT‑5 generated a fully functional French-learning website—complete with interactive games and polished frontend—based on a single paragraph prompt. Currently, tech titan Microsoft has already begun including GPT-5 into its products and services. This includes Microsoft Copilot, whose reasoning capabilities will get a boost, and Microsoft notes that “customers with Microsoft 365 Copilot licenses can also reason over emails, documents and files.”
In addition to this, OpenAI spent more than 5,000 hours testing GPT‑5 to reduce hallucinations. The model now supports “safe completions,” providing constrained yet helpful responses to sensitive queries rather than outright refusals. Alex Beutel, head of safety research, emphasized GPT‑5’s improved capacity to admit when it’s unsure—a step forward in building user trust. While stopping short of calling GPT‑5 AGI, Altman described it as “clearly a model that is generally intelligent,” though still missing real-time learning capabilities.
The launch comes a few hours after an accidentally published blog post on GitHub had spilled the beans on GPT‑5. Though quickly deleted, the post had already been archived and widely shared online, offering an early and extensive look into the company’s newest advanced AI system to date. Screenshots from the leak suggest GPT‑5 is not a single model – instead, it comes with a suite of variants that can serve a range of use cases across consumer and enterprise apps.
The timing of this latest launch becomes more crucial, as recent reports suggest that Microsoft is in talks with OpenAI to revise the AGI clause and extend access to advanced technologies. Actually, tensions between OpenAI and its prominent backer, Microsoft, have been growing over recent years. The concerns mainly revolve around OpenAI’s shift to a for-profit model and disagreements over revenue sharing.
