Codex usage limits are hard to compare because OpenAI now describes some plans in estimated messages, others in allowance or credits, and API-key use in API billing. A “message count” isn’t a reliable quota for every task.
Short answer: Check your current balance and reset time in ChatGPT under Settings → Usage; in an active Codex CLI session, enter /status. As of October 2026, Plus and Business Standard have published five-hour message estimates, Pro has no five-hour cap but still has an included allowance, and Business Premium has a higher allowance with no five-hour cap. Enterprise/Edu depends on workspace billing; an API key is billed separately at API rates.
What are Codex usage limits by plan?
Codex usage is an allowance, not a guaranteed count of prompts. I checked the table against OpenAI’s current Codex pricing and usage page and plan-specific guidance on October 4, 2026. OpenAI can change the figures; your Usage page is the live source for your account.
| Plan or sign-in | What the published limit looks like | What to check |
|---|---|---|
| Plus | Estimated local messages per five-hour window: GPT-6 Astra 5–45; GPT-6.1 Sol 15–160; GPT-6 Sol 15–150; GPT-6 Luna 350–3,000. A weekly limit may also apply. | Usage page for remaining amount and reset time. Estimates vary by task and settings. |
| Pro 100, Pro 200, Pro 500 | No five-hour Work/Codex limit as of this check. An included allowance still applies, and exact usable work depends on model and task. | Usage page; do not read “no five-hour limit” as unlimited use. |
| Business Standard | Published local-message estimates match Plus for Astra, GPT-6.1 Sol, GPT-6 Sol and Luna. A weekly limit may also apply. | Confirm your seat and workspace controls with the admin. |
| Business Premium | Five times the Standard seat’s included usage and no five-hour limit. | Usage page and workspace credit/spend controls. |
| Enterprise / Edu | Flexible-pricing workspaces have no fixed rate limit; use scales with workspace credits. Without flexible pricing, per-seat usage matches Plus for most features. | Ask the admin which billing arrangement and limits apply to your workspace. |
| API key | Uses API model availability and API billing, not the ChatGPT plan allowance. API-key use does not include Codex cloud features. | Check the API Usage Dashboard and your organization’s Limits page. |
OpenAI labels the Plus and Business Standard ranges as estimates; cloud tasks may use more allowance than local messages. There’s also a small documentation mismatch: the general pricing page lists GPT-6.1 Sol at 15–160 local messages for Plus and Standard Business, while the Business models page lists GPT-6.1 Sol and GPT-6 Sol together at 15–150 for Standard seats. For plans with both a five-hour and weekly window, either of those Codex limits can stop new work. “Weekly” also doesn’t mean the same thing as an API rate limit or monthly workspace budget.
How do I check Codex usage and find its reset time?
Here’s how to check Codex usage: use the account Usage page for a live plan balance and reset; use /status when you want a reading from the active session. Codex use from the CLI, IDE extension, desktop and web shares the allowance attached to the signed-in plan, so changing clients doesn’t create a fresh pool.
- Web: Open ChatGPT, select your profile, then Settings → Usage. Check which limit is shown, what remains, and the reset time.
- CLI: In the active Codex session, type
/status. The result includes available rate-limit information. The slash command and the Codex usage dashboard are documented in OpenAI’s Codex plan guide. - IDE extension: Check the same ChatGPT Settings → Usage page for the account’s shared allowance. If you’re in a Codex desktop chat instead, enter
/statusin the composer; it reports rate limits.
Save the Codex usage reset time with its time zone before planning a long task. Where the Codex 5-hour limit applies, OpenAI says its window begins when you send your first Work or Codex message after the previous window ends. The Codex weekly limit has a separate window. Trust the reset displayed for your account rather than estimating from midnight, the date of your first subscription payment, or someone else’s screenshot.
What uses the Codex allowance?
The model, reasoning and speed settings, task complexity, context, tools, and whether work runs locally or in the cloud all affect usage. OpenAI says two tasks that look similar can consume different amounts, so prompt length alone isn’t a usage calculator.
Spot a costly pattern: run comparable tasks with different model and reasoning settings and check the Usage page after each. In eligible Enterprise workspaces on credit-based billing, members can also see their monthly allowance, recent credit history, their highest-usage chats and each chat's lifetime credit use. On other plans, use the visible balance and reset indicators.
Cloud isn’t automatically cheaper than local. OpenAI’s estimates say cloud tasks may use more of the allowance than local messages, and task size and context still matter. A broad request that asks Codex to inspect a large repository, keep a long history, call many tools and produce several optional deliverables has more usage drivers than a targeted change with a small set of relevant files.
How can I make the usage last longer?
Reduce repeated context and right-size the task before changing plans. OpenAI’s own guidance recommends concise instructions, relevant source files, a defined output, smaller project instructions and fewer enabled MCP servers.
- State the outcome and acceptance check in the first prompt. Split independent work instead of adding optional improvements to the same run.
- Point Codex at the files and date range that matter. Avoid attaching the entire repository or repeating the same long background in every message.
- Match the model and reasoning setting to the risk: use a lighter option for routine edits if it still meets your review standard, and keep deeper reasoning for work that needs it. Check your Usage page after a comparable task; there is no reliable universal savings percentage.
- Trim
AGENTS.mdinstructions and disable MCP servers you don’t need for that task. OpenAI says these add context to messages. - For repeated edits to one file, batch the operations if your tools support it. We build octofs at Muvon; its
batch_edittool can apply up to 50 same-file, non-overlapping edits atomically and return fresh line IDs without a re-view. That can remove tool round trips from this particular workflow; it doesn’t alter OpenAI’s billing rates or prove a specific token saving. See the Octofs 0.16 guide.
What happens when I hit the limit?
The options depend on your plan and the limit notice. If a turn is already running, OpenAI says it may continue to completion subject to fair-use limits; after that, the notice or Usage page may offer credits, an eligible reset, an upgrade, or a wait until the displayed reset.
Plus and Pro users can buy Codex credits where the option is available. Business, Edu and Enterprise workspaces on flexible pricing can use workspace credits if the balance and admin spending controls allow it.
A local Codex session can also use an API key, but that moves usage to separate API rates and doesn’t add Codex cloud features. API request/token rate limits, organization usage limits, project spend limits and prepaid balance are separate controls. If you choose this route, check the API Usage Dashboard and API rate-limit and 429 guidance first; our guide to running one AI agent across many models explains the routing trade-off.
Don’t treat an old error string as a current guarantee. A 2025 GitHub report from Codex CLI 0.39.0 showed “You've hit your usage limit. Upgrade to Pro … or try again in 4 days 6 hours 37 minutes.” The wording and timing are historical. Today, compare the specific limit banner with Settings → Usage and the reset time it shows. If the balance looks wrong after that time, send Support the screenshot, client, model, time and time zone; OpenAI says Support can’t manually reset a limit.
The useful number is the one your account shows now. Check the allowance before a long job, then use its model, reasoning, task history and context size to decide what to keep, narrow or move to separately billed API usage.
— Don
Octofs is open source. If you try batch_edit in your Codex sessions, open an issue with what worked and where it got in the way.
