GPT-5 Codex learns when to think fast — and when to think deep

OpenAI’s new GPT-5 Codex can now decide how long to “think” on coding tasks: snapping back quick answers for simple queries, or spending up to 35 minutes coding through complex refactors.

‍

- Dynamic reasoning: Short tasks get instant replies, while tricky planning prompts trigger deeper, structured thinking.

- Seamless workflows: Work can shift between local VS Code and Codex Cloud, with context preserved even if you shut your laptop.

- Smarter reviews: A repo-running bot now executes tests, applies fixes, and pushes changes directly on GitHub.

‍

Codex is edging closer to a true coding partner — though multi-sprint projects still stretch its limits.