OpenAI launches GPT-5, its most advanced AI model yet

On August 7, 2025, OpenAI announced GPT-5, the latest and most capable model in its ChatGPT product line. Positioned as a major leap over GPT-4o and other predecessors, GPT-5 delivers state-of-the-art performance across coding, mathematics, writing, health, and multimodal reasoning.

Unified and adaptive reasoning

GPT-5 introduces a unified system that automatically selects between rapid responses and deeper reasoning (“GPT-5 Thinking”) based on task complexity and user intent. Paid Pro subscribers gain access to GPT-5 Pro, an extended-reasoning variant achieving top scores in challenging benchmarks like GPQA and SWE-bench.

Key improvements

1. Reduced hallucinations

~45% fewer factual errors vs GPT-4o, and up to 80% fewer than OpenAI o3 in reasoning mode.

2. Enhanced coding

Stronger front-end generation, large-repo debugging, and improved design sensibility.

3. Advanced writing

More nuanced literary style, improved handling of free-form and ambiguous structures.

4. Health expertise

Top performance on HealthBench Hard, with proactive and context-aware answers.

5. Multimodal reasoning

State-of-the-art results in visual, video, and spatial understanding tasks.

Availability

GPT-5 is rolling out now as the default model for all signed-in ChatGPT users. Free users will have limited usage before switching to GPT-5 mini, while Plus, Pro, Team, Enterprise, and Edu plans enjoy higher limits and GPT-5 Pro access.

OpenAI says GPT-5 represents a step toward more accurate, honest, and safe AI, with new safe completion training reducing unnecessary refusals and balancing safety with usefulness.