OpenAI Codex cuts max context from 372k to 272k tokens
Original: OpenAI reduces Codex Model Context Size from 372k to 272k
Why This Matters
A 27% context window reduction in a widely-used coding AI tool directly impacts developer workflows and task complexity limits.
OpenAI merged a pull request on July 18, 2026 to update bundled model metadata in Codex CLI version 0.144, reducing the Codex model's maximum context window from 372,000 tokens to 272,000 tokens.
OpenAI engineer sayan-oai merged pull request #33972 into the openai/codex repository on July 18, 2026, targeting the release/0.144 branch. The single-commit change is described as a 'backport of refreshed bundled model metadata' and modifies one JSON file. The key change reduces the maximum context size for the Codex model from 372,000 tokens to 272,000 tokens — a reduction of 100,000 tokens. No public explanation was provided in the PR for the reason behind the decrease. The openai/codex repository, which serves as the codebase for the Codex CLI tool, currently has 99,700 stars and 14,900 forks on GitHub.