What the ChatGPT 5? Is it free, what's new, how to get it…all questions answered
OpenAI has released GPT-5 to all ChatGPT users, introducing new model variants, broader system integrations, and developer pricing.
OpenAI has started rolling out GPT-5, its latest language model, to both free and paid ChatGPT users. The update introduces new variants, broader system integrations, and improvements in performance and safety. So, how do you get to use ChatGPT 5? Well, it’s simple, you will automatically be shown the new ChatGPT-5 welcome screen. It’s a staged roll out so it may take a while for everyone to be on GPT-5. However, you can always try to relogin to check.
GPT-5 is being positioned as a step forward in OpenAI’s long-term pursuit of artificial general intelligence. While the model does not meet AGI benchmarks as defined in OpenAI’s own charter, it is described internally as generally intelligent. It does not have the ability to learn after deployment.
The company claims GPT-5 is more accurate and efficient, with reduced hallucination rates compared to earlier versions. OpenAI’s CEO Sam Altman said the leap from GPT-4 to GPT-5 is comparable to the display upgrade from early iPhones to Retina displays.
Two new model variants have been launched: GPT-5-mini, a lightweight version, and GPT-5-nano, a faster and cheaper model available only via API. Free users will have access to GPT-5 and GPT-5-mini. Paid tiers unlock higher usage limits, with the $200/month Pro tier offering GPT-5-pro, GPT-5-thinking, and unlimited access.
The ChatGPT interface now auto-selects the most suitable model based on query complexity and user plan. Legacy models remain accessible for Pro users.
On the API front, GPT-5 is priced at $1.25 per million input tokens and $10 per million output tokens. GPT-5-mini is priced at $0.25 per million input and $2 per million output tokens. GPT-5-nano is priced at $0.05 per million input and $0.40 per million output tokens. The nano variant is now priced lower than Google’s Gemini 2.5 Flash.
From next week, Pro users will be able to integrate Gmail, Google Calendar and Contacts with ChatGPT. Other tiers will receive access later. ChatGPT will reference these accounts automatically when relevant, without manual selection.
Users can now select from four preset personalities: Cynic, Robot, Listener, and Nerd. These modes are also planned for integration with Advanced Voice Mode. The chat colour can also be customised.
The API allows toggling between detailed and direct responses. GPT-5 features a 256,000-token context window, up from 200,000 in the previous o3 model. This enables better retention across long conversations, code and documents.
GPT-5 outperforms earlier models in coding tasks. Benchmarks include 74.9 percent on SWE-Bench Verified, 55 percent on SWE-Lancer using GPT-5-thinking, and 88 percent on Aider Polyglot. These cover bug fixing, multi-language support and freelance-style tasks.
During a demo, OpenAI’s team instructed GPT-5 to build an interactive language-learning app. The system generated the requested features and UI within a minute. OpenAI claims the model can handle complex task chains and tool use, including browser and API-based calls.
In health benchmarks, GPT-5-thinking scored 25.5 percent on HealthBench Hard, compared to o3’s 31.6 percent. OpenAI states that its new model performs better across health-related tasks, with outputs reviewed by medical professionals.
The company reports that hallucination rates have declined. GPT-5-thinking shows a 65 percent lower rate compared to o3, and 26 percent lower than GPT-4o. Safety updates include reduced chances of the model producing deceptive, hacked or unsafe outputs. Dual-use prompts are handled with what OpenAI calls "safe completions", offering responses within controlled boundaries. Over 5,000 hours of red teaming and external testing were conducted to improve robustness.
The system is designed to fail safely when faced with unsolvable prompts. This includes refusing tasks beyond its capabilities without generating misleading outputs.
OpenAI reports 700 million weekly ChatGPT users, 5 million paying business users, and 4 million developers on the API.
What’s new in GPT-5: Key changes at a glance:
- Available to All Users: GPT-5 is being rolled out to both free and paying ChatGPT users.
- New Model Variants:
- GPT-5-mini (lightweight)
- GPT-5-nano (API only, faster and cheaper)
- GPT-5-pro and GPT-5-thinking (for Pro-tier users)
- Automatic Model Selection: The chat interface now auto-selects the model based on the complexity of the query and user tier. Manual switching is no longer required.
- Pro-Tier Benefits:
- Unlimited GPT-5 access
- Access to GPT-5-pro and GPT-5-thinking
- Legacy models still available
- Google Integration: Pro users can soon connect Gmail, Google Contacts, and Google Calendar to ChatGPT. Other tiers will get this later.
- Customisation Features:
- Users can choose chat colours
- Four built-in personality modes: Cynic, Robot, Listener, Nerd
- Will be integrated with Advanced Voice Mode later
- Performance Boosts:
- Context window increased to 256,000 tokens (up from 200,000 in previous o3 model)
- Lower hallucination rate than GPT-4o and o3
- Improved performance in coding benchmarks like SWE-Bench, SWE-Lancer, and Aider Polyglot
- Better at agentic tasks and tool execution (e.g., API and browser usage)
- Developer Pricing:
- GPT-5: $1.25/1M input tokens, $10/1M output tokens
- GPT-5-mini: $0.25/1M input, $2/1M output
- GPT-5-nano: $0.05/1M input, $0.40/1M output
- Health-related Performance:
- GPT-5-thinking outperforms o3 in health benchmarks
- Claimed to be OpenAI’s best model for medical questions
- Evaluated by physicians; designed to fail gracefully on unsolvable tasks
- Safety Improvements:
- 26% reduction in hallucinations over GPT-4o
- 65% reduction over o3 for GPT-5-thinking
- Safer completions for dual-use prompts
- 5,000+ hours of red teaming and external safety testing
- Reduced likelihood of deceptive or unsafe behaviour
