OpenAI has unveiled GPT-5, its most advanced AI model to date, featuring an adaptive system that automatically switches between fast responses and deep reasoning based on query complexity. Available to all ChatGPT users, including free-tier, GPT-5 boasts significantly reduced factual errors—45% fewer than GPT-4o—and an 80% error reduction in reasoning mode compared to previous versions.
GPT-5 excels across multiple domains: it scored 74.9% on coding benchmarks, demonstrating strong capabilities in generating full software applications; improved health-related query accuracy with a 46.2% score on HealthBench Hard; and achieved 94.6% on the AIME 2025 math test. It also surpassed average human performance in reasoning tasks, scoring 90% on SimpleBench against an 83% human average.
The model uses a real-time routing system to select appropriate computing modes and incorporates new "Safe Completions" safety features that prioritize helpful and secure responses without outright refusals. Usage limits vary by subscription level, with free users accessing GPT-5 until reaching caps and Pro users enjoying unlimited access and enhanced features. Enterprise and education users will receive access soon.
