OpenAI has officially released GPT-5, describing it as its “smartest, fastest, most useful model made”. The new artificial intelligence model is now available to all ChatGPT users, including those on the free plan, albeit with certain usage limits. This launch marks a significant advancement in OpenAI’s suite of language models, promising enhanced reasoning, improved performance, and a better overall user experience.
GPT-5 and its companion models
Alongside the main release, OpenAI has introduced two companion models: GPT-5-mini and GPT-5-nano, both made available on its API platform for developers. While the full GPT-5 and GPT-5-mini are generally accessible to users, the nano version is designed for very lightweight tasks and internal use cases.
The key distinction between GPT-5 and GPT-5-mini lies in their performance and efficiency. GPT-5 is designed for more complex tasks, featuring deeper reasoning and broader capabilities, making it well-suited for professionals, creatives, and researchers. In contrast, GPT-5-mini is a smaller, more cost-effective model built for routine tasks where speed and simplicity are more important than depth. As such, GPT-5-mini may omit some nuance or detailed explanation to remain fast and concise.
GPT-5 introduces a “unified system” architecture, allowing users to interact with a single model that automatically switches between underlying tools and components depending on the task at hand. For instance, GPT-5 utilises a deeper reasoning model, called GPT-5 Thinking, when needed, but can switch to a real-time router for faster responses. The model also adapts based on user preferences, such as whether a person favours quick answers or more detailed, thoughtful ones.
Improvements in reasoning, writing, and reliability
OpenAI reports that GPT-5 significantly outperforms its predecessors across a range of benchmarks. It is faster, more reliable, and better at following instructions. The model has been refined to reduce hallucinations—false or misleading information often generated by AI systems. It also demonstrates enhanced accuracy in responding to health-related queries, along with stronger capabilities in writing and coding.
Notably, GPT-5 has been trained to avoid excessive sycophancy, meaning it is less inclined to agree with users to appear helpful or agreeable. This change is intended to make responses more grounded and balanced, improving trust and overall quality.
In writing tasks, GPT-5 shows improvements in generating emotionally compelling content. A demonstration showed it outperforming GPT-4 in producing poetry with a clearer emotional arc, richer imagery, and more vivid metaphors. These enhancements also extend to more practical tasks, including report drafting, email composition, and editing assistance, making the tool more valuable for everyday users.
Custom personalities and access tiers
In addition to technical upgrades, OpenAI is experimenting with a research preview of four customisable personalities for ChatGPT. These include the Cynic, Robot, Listener, and Nerd—each offering a different tone or style of interaction. These personalities can be enabled or disabled in settings and are initially available for text-based chats. Voice support is expected to follow.
As with previous releases, access to GPT-5 is divided between free and paid users. All ChatGPT users, including those on the free plan, can now try GPT-5. However, usage is capped for free users; once that limit is reached, the system automatically switches to GPT-5-mini.
Subscribers to the Personal plan (ChatGPT Plus) enjoy significantly higher usage limits and can use GPT-5 as their default model. Pro subscribers also gain access to GPT-5 Pro, which provides the most robust experience. For business users under Team and Enterprise plans, GPT-5 is available as a default model with usage limits designed to support full organisational workflows.
