Keep pulling the thread on Noam Brown.
OpenAI launched its GPT-5 model system in August 2025.
The GPT-5 system is a unified architecture composed of a 'smart and fast model' for most questions and a 'deeper reasoning model' for more difficult problems.
The GPT-5 system outperforms previous OpenAI models on benchmarks.
GPT-5 has significantly reduced hallucinations compared to previous OpenAI models.
GPT-5 has improved instruction-following capabilities compared to previous OpenAI models.
All GPT-5 models incorporate a safety training approach called 'safe-completions' to prevent the generation of disallowed content.
OpenAI has classified the gpt-5-thinking model as having 'High capability' in the Biological and Chemical domain under its Preparedness Framework.
GPT-5 uses a real-time router to dynamically select the appropriate model for a query based on conversation type, complexity, tool needs, and explicit user intent.
The router for GPT-5 is continuously trained on user signals, including model switching, response preference rates, and measured correctness, to improve its performance over time.
The GPT-5 system answers questions more quickly than previous OpenAI models.
GPT-5 has minimized sycophancy compared to previous OpenAI models.
GPT-5 has improved performance in writing tasks compared to previous models.