OpenAI quietly shrank the context window of Codex — its agentic coding tool that reads your repo and ships PRs — by 27%. No changelog, no announcement. Developers spotted the change in the open-source Codex CLI on GitHub, and the thread hit the Hacker News front page with 273 points and 130 comments.
Why 100k tokens matter
Codex’s whole pitch is long-horizon agentic coding: load a big codebase, let the agent work. Cutting 372k to 272k means smaller repos fit and long sessions compact earlier. Same subscription price, less room. OpenAI’s defense: requests above 272k input tokens bill at 2x, so the cap “protects your quota.” Users read it differently — the advertised long-context tier is now unreachable inside Codex, and some report the effective window dropping to around 258k.
The API angle
Codex runs on the OpenAI API and ChatGPT subscriptions, so anyone building coding agents on GPT-5.6 inherits the same ceiling. The timing is rough: this landed right after OpenAI encrypted Codex’s sub-agent prompts and kept resetting usage limits — a tracker site mocking those resets scored 287 points on HN the same day. Trust in the black box is thinning.
You Might Also Like
- Codex Starts Encrypting sub Agent Prompts Openai Codex 28058 and you Cant Read Your own Agent Logs Anymore
- Openai Codex Pets Turn Your ai Coding Agent Into a Desktop Tamagotchi
- Openai Codex in Chrome Moves the Coding Agent Into Your Real Browser Session
- Openai Codex Record and Replay Demo a mac Task Once the Agent Repeats it Forever
- Google Gemini 3 5 pro ga Ships With a 2m Token Context Window the Biggest in any Production Model
Top AI Product 
Leave a comment