The Codex context window is not your usage limit

The Codex context window is how many tokens the active model can hold in one chat: your instructions, the files it has read, tool output and the conversation so far. It is not the same thing as your Codex usage limit, and mixing the two up is the most common reason people chase the wrong problem.

You can check how full the context is from inside Codex. The size itself depends on the model, which OpenAI changes often, so this piece does not quote a number.

Context window and usage limit are different clocks

A chat can run out of context while your plan still has plenty left, and the reverse happens too.

How to check the Codex context window

In the Codex CLI, type /status. OpenAI's docs say it shows the active model, approval policy, writable roots and remaining context capacity, along with token usage. The footer of the terminal UI also shows how much context is left as a percentage while you work, and /statusline lets you choose which fields appear there.

If you run Codex through a config file, model_context_window in config.toml sets the context window tokens available to the active model. Most people leave it alone.

What happens when the context fills

Codex does not simply stop. Once the chat passes a token threshold, it compacts the history automatically, summarizing older turns so the work can continue. The threshold uses model defaults unless you set model_auto_compact_token_limit.

You can also do it yourself. /compact summarizes the visible chat to free tokens, and OpenAI suggests it after long runs. /new starts a fresh chat in the same CLI session, which empties the context completely. A summary keeps the key points but drops detail, so for a new task a fresh chat is usually the cleaner choice.

Why a full context costs more than space

Every turn sends the chat back to the model. A context that is close to full means each new message carries a large history with it, and that history counts as tokens against your usage. OpenAI's own tips for making a usage allowance last longer point the same way: smaller prompts, fewer source files, a shorter AGENTS.md and fewer MCP servers. Each one also leaves more room in the context window.

Watching the other clock

The context window is per chat and lives inside Codex. Your usage over days and weeks is a separate question. UsageAtlas, a free and open-source desktop tracker, covers that side: Codex session and weekly windows, reset times, and token history by day and model from your local Codex activity. It does not show the context window size, and it says so plainly, so keep /status for that.

If you only work in short chats and never hit a limit, /status is all you need. A tracker earns its place when you want to know why a week ran out early and which model or project used the tokens.

The short version

Context is working memory for one chat. The usage limit is your plan's allowance over time. Check context with /status, manage it with /compact and /new, and keep an eye on the allowance with the usage dashboard or a desktop tracker such as UsageAtlas for Codex.