GPT-5 is now default in ChatGPT - what changes first

GPT-5 rollout details

The company calls GPT-5 its most accurate model to date. The global switch started at 09:00 PT and is scheduled to finish within 24 hours. Enterprise and Education accounts gain access next week.

• Free, Plus, Pro, and Team accounts can use GPT-5 immediately. After daily limits, free users are redirected to a "GPT-5 mini" variant.
• Plus and Team subscribers retain higher limits, while Pro users unlock a "GPT-5 Pro" setting optimised for large code and data tasks.
• Enterprise administrators can enable GPT-5 from 20 May 2025 via the admin console.

OpenAI released GPT-5 directly to the free tier for the first time, citing a goal of "factual answers for everyone."

Broader access for ChatGPT users

• Evaluations show GPT-5 produces 45% fewer factual errors than GPT-4o.
• Error reduction reaches 80% when users activate deeper reasoning mode.
• The model now accepts multimodal prompts combining text, images, and charts without extra plug-ins.
• A revised safety layer favours partially helpful replies over blanket refusals when content nears policy limits.

Background

GPT-4o launched in April 2024 with faster responses but faced criticism for hallucination rates. OpenAI focused GPT-5 on source-grounded reasoning, running a pilot with 5,000 external developers from February to April 2025. Regulators in the EU and United States received advance briefings under existing AI oversight frameworks.

Sources

• OpenAI press release, "GPT-5: Accuracy and Transparency," 13 May 2025.
• Technical report, "Introducing GPT-5," 13 May 2025.