We study how training on incorrect responses can cause broader misalignment in language models and identify an internal feature driving this behavior—one that can be reversed with minimal fine-tuning.

We are replacing the existing GPT-4o-based model for Operator with a version based on OpenAI o3. The API version will remain based on 4o.

Codex is a cloud-based coding agent. Codex is powered by codex-1, a version of OpenAI o3 optimized for software engineering. codex-1 was trained using reinforcement learning on real-world coding tasks in a variety of environments to generate code that closely mirrors human style and PR preferences, adheres precisely to instructions, and iteratively runs tests until passing results are achieved.

HealthBench is a new evaluation benchmark for AI in healthcare which evaluates models in realistic scenarios. Built with input from 250+ physicians, it aims to provide a shared standard for model performance and safety in health.

OpenAI o3 y OpenAI o4-mini combinan el razonamiento más avanzado con todas las funcionalidades de la herramienta: navegación web, Python, análisis de imágenes y archivos, generación de imágenes, lienzo, automatizaciones, búsqueda de archivos y memoria.

Sharing our updated framework for measuring and protecting against severe harm from frontier AI capabilities.

Introducing GPT-4.1 in the API—a new family of models with across-the-board improvements, including major gains in coding, instruction following, and long-context understanding. We’re also releasing our first nano model. Available to developers worldwide starting today.

BrowseComp: a benchmark for browsing agents.