OpenAI just launched GPT-5, calling it their most advanced AI system yet with significant improvements in coding, math, writing, and health applications.
The unified system automatically decides when to use quick responses versus deeper reasoning based on query complexity.
GPT-5 demonstrates major performance gains, achieving 94.6% on AIME 2025 math problems and 74.9% on coding benchmarks.
The model addresses key AI limitations by reducing hallucinations by 45% compared to GPT-4o and cutting sycophantic responses in half.
It introduces “safe completions” training for more nuanced safety responses rather than simple refusal.
GPT-5 replaces previous models as ChatGPT’s default, with availability rolling out to all user tiers.
source