OpenAI has launched GPT-5, its most advanced language model to date, and made it available for free to all ChatGPT users. Announced by CEO Sam Altman during a livestream, the upgrade brings major improvements in reasoning, accuracy, memory, personalization, and integration with real-world tools.
GPT-5 introduces “agent-like” thinking, meaning it can break down complex requests, understand context, and act more independently to help users. It also makes far fewer mistakes compared to earlier models, with hallucination rates dropping to 4.8% from over 20% in GPT-4o. In healthcare-related tests, error rates were reduced to just 1.6%.
New features include “vibe coding,” which allows developers to describe the mood or style of a project and receive fully functional code. GPT-5 also supports different personalities such as Cynic, Listener, Nerd, and Robot, allowing users to choose a tone that suits their style. Voice Mode has been upgraded to sound more natural, with adaptable pacing and tone.
The model comes in three sizes, GPT-5, GPT-5-mini, and GPT-5-nano (API only). Free users get access to the main GPT-5 model with all core capabilities, while Pro subscribers get unlimited usage and access to GPT-5 Pro and GPT-5 Thinking for deeper, more tool-assisted responses.
Pro users can also link GPT-5 with Google services like Gmail, Calendar, and Contacts for scheduling, drafting emails, and managing tasks automatically. OpenAI says it only accesses data when necessary.
On benchmarks, GPT-5 leads in several areas, including coding and PhD-level science questions, outperforming competitors like Claude Opus 4.1 and Gemini 2.5 Pro in many categories. Safety has also been improved, with stronger detection of bad-faith prompts and more transparency when uncertain.
With this release, OpenAI is aiming to make powerful AI tools more accessible to everyone, not just premium users, bridging the gap between a chatbot and a truly intelligent digital assistant.
