OpenAI shrank the Codex context window on purpose

Canonical version: OpenAI shrank the Codex context window on purpose.

OpenAI reduced the Codex model's context size from 372k to 272k tokens. There was no announcement; the change appeared in a pull request in the open-source Codex CLI repo, and that pull request reached the Hacker News front page.

Thibault Sottiaux (Codex team) explained the reason on X: cache reads get more expensive as a bigger context is sent back and forth between tool calls, so the cheapest setting isn't the maximum length. According to him, output quality is similar above 272k; the gains of a bigger window are speed (no waiting for compaction) and handling very large inputs. OpenAI is working on tuning that would let them raise the limit again without charging more usage.

I think you should Treat AI context as a budget to manage. Spend tokens on what the current task needs, keep durable knowledge in specs and instruction files outside the window, send exploration to subagents, and end long runs deliberately. OpenAI builds the model and chose the smaller window for Codex; I'd follow their lead.

Simon Willison noticed that the same commit adds system-prompt guardrails telling Codex to resolve targets with read-only checks before destructive actions, and never to point a recursive delete at $HOME or /.

References

• Hacker News discussion: https://news.ycombinator.com/item?id=48965850
• The PR: https://github.com/openai/codex/pull/33972/files
• Explanation from Thibault Sottiaux: https://x.com/thsottiaux/status/2076543065045795309

Related

• Context size reduction
• Treat AI context as a budget to manage
• Context Engineering

About Sébastien

Ready to get to the next level?