Insights on evaluating large language models

GPT-5 is a unified system with a smart and fast model that answers most questions.
OpenAI[1]
We’ve made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy.
OpenAI[1]
safe-completions seek to maximize helpfulness subject to the safety policy’s constraints.
OpenAI[1]
gpt-5-thinking generally performs on par with OpenAI o3.
OpenAI[1]
Our data processing pipeline includes rigorous filtering to maintain data quality.
OpenAI[1]
Space: Let’s explore the GPT-5 Model Card