OpenAI reduces Codex Model Context Size from 372k to 272k
Points and comments are a snapshot, not live.
OpenAI reduced Codex's context size from 372k to 272k tokens.
OpenAI merged a pull request backporting refreshed model metadata to Codex version 0.144. The change reduced the model context size from 372k to 272k tokens. The commit also added a new system prompt section that tells the model to avoid destructive actions like deleting files in broad directories.
What commenters are saying
Commenters largely saw the context cut as a cost/latency tradeoff, not a capability issue, noting most sessions don't need the full 372k. OpenAI's Tibo stated it was temporary and would return to 372k soon. Some users praised Codex's compaction system for handling large contexts well, though others reported it occasionally forgets tasks during compaction. A few pointed to DeepSeek's superior caching as a better model for cost efficiency. Several commenters debated the merits of open-sourcing Codex.