Analysis · Codex limits · 3 min read
Codex usage limits: resets and a task budget
A practical worksheet for comparing completed work inside the same usage window.
Codex limits are easiest to understand when you separate the allowance in your subscription, the reset window, and the cost of the task you are running. This guide gives you a way to diagnose a limit and compare workflows without treating a message count as a promise. Checked on 1 November 2026.
Find the limit that stopped you
OpenAI's official pricing documentation points to the usage dashboard and the CLI's /status command for your remaining allowance and reset times. Plus has a five-hour window; Pro currently has no five-hour limit. Weekly limits can still apply. Local messages and cloud chats share the allowance. API-key usage is billed separately.
Before retrying, record the plan, sign-in method, model, speed mode and the exact limit shown. A weekly limit and a short session limit call for different decisions. Use the reset time displayed for your account rather than a time copied from someone else's screenshot.
Measure completed tasks
A useful comparison starts with one deliverable: a passing bug fix, an accepted document, or a reviewed change. Define what counts as success before trying another workflow.
| Record | Why it matters |
|---|---|
| Starting and ending allowance, with time | Keeps the comparison inside one reset window |
| Model, speed and reasoning setting | Makes the run repeatable |
| Task and acceptance check | Prevents a shorter but incomplete answer from winning |
| Accepted results, retries and elapsed time | Includes the work needed to repair a result |
For an illustrative comparison, workflow A uses 20 percentage points for two accepted results: 10 points per accepted result. Workflow B uses 25 points for five: five points per result. These are invented worksheet numbers, not Codex benchmarks. Use comparable tasks and the same allowance; a rounded meter makes small differences unreliable.
Change one thing at a time
Start with a bounded task and a concrete check. For a code change, name the behavior to change and the command that verifies it. Keep the repository instructions available, and give the agent the relevant error or reproduction.
For a comparison, hold the task set and acceptance criteria constant. Change one choice, such as model or speed, then record the outcome. Include failed attempts. Repeating only the successful case gives an optimistic budget.
Do not extrapolate one short run into a guaranteed day of work. A repository investigation and a one-file edit have different workloads. Keep separate rows for routine work and the large tasks that consume the expensive tail.
What to do at the limit
Save the current result and the exact next step. Decide whether the task can wait for the displayed reset, needs a smaller scope, or merits a separately budgeted continuation. A useful handoff says what changed, what passed, what failed and where to resume.
Use the Codex setup, pricing and limits guide for the starting workflow, dated announcements and temporary promotions. For monetary API budgeting, use the API cost worksheet. A historical reset announcement is useful context, but the account dashboard is the place to check whether you have capacity now.
If you are considering another coding subscription, the Codex, Claude Code and Cursor cost comparison separates included usage from extra charges. Compare the cost of an accepted task before deciding whether to upgrade or switch tools.
Keep the reference current
Recheck the official page when the product changes, then update the facts and the review date together. Keep promotion dates with the announcement rather than turning them into permanent plan features. This page explains how to check and measure usage; it does not predict the allowance of an individual account.