OpenAI Reduces Codex Model Context Size
OpenAI has quietly reduced the context size of its Codex AI model, according to a recent GitHub pull request. The change lowers the maximum context window from 372,000 tokens to 272,000 tokens.
The pull request, numbered 33972, was merged on July 18, 2026. It was authored by a developer with the handle sayan-oai and merged into the release/0.144 branch of the openai/codex repository on GitHub.
Details of the Change
The commit message reads "Backport refreshed bundled model metadata to 0.144." The pull request itself is titled "Backport refreshed bundled model metadata to 0.144." The change involves 64 additions and 54 deletions across one file.
The affected file is a JSON file containing model metadata. The pull request was part of a hotfix for version 0.144 of Codex.
Context and Background
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Codex is OpenAI's AI system that translates natural language into code. It powers GitHub Copilot and other developer tools. The model's context size determines how much text the AI can consider at once when generating responses.
The reduction from 372,000 to 272,000 tokens represents a decrease of approximately 27 percent. Tokens are pieces of text that the model processes; one token is roughly equivalent to four characters in English.
Community Reaction
The change was noted by developers on Hacker News, where the pull request was shared. Some users expressed concern about the reduced capacity, while others noted that the change might improve performance or reduce costs.
OpenAI has not publicly commented on the reason for the reduction. The pull request does not include a detailed explanation of the change.
The repository remains public on GitHub with 99,600 stars and 14,900 forks. It has over 5,000 open issues and 240 pull requests.