OpenAI Reduces Codex Model Context Size from 372k to 272k
OpenAI reduces Codex Model Context Size from 372k to 272k
I recently merged a pull request to backport refreshed bundled model metadata for OpenAI Codex version 0.144. This update specifically reduces the model context size from 372k to 272k, ensuring the release branch stays synchronized with the latest configuration changes. The modification involves updating a single JSON file to reflect these critical adjustments for the AI code creation tool.
"Backport refreshed bundled model metadata to 0.144"
HN discussion
- Practitioners argue that models perform worse with excessive context, recommending breaking work into modular 300k token chunks to maintain accuracy and reduce costs rather than relying on massive windows.
- Users are actively migrating from OpenAI Codex to Claude Code, citing Claude's 1M token context limit and superior compaction handling as decisive factors for their workflow.
- Developers advocate for hierarchical planning where high-level architecture and decisions are written to persistent files, allowing the AI to reference specific documents instead of stuffing the entire history into the context window.
- Tools like Pi and custom extensions address compaction gaps by visualizing 'compaction capsules' or using skills like /total-recall to retrieve pre-compaction conversation history on demand.