Codex limits: why they run out and how to keep working

You were working with Codex, and at some point it said the usage limit was exhausted. Questions usually follow: how to reset the Codex limit, is there a free way, what to check in usage. Below is a calm breakdown: what drives usage, why there is no "reset", where to see what is left and how to continue on your own key with a dollar budget.

The subscription details come from OpenAI's help and documentation as of October 18, 2026. They change, so for exact numbers for your plan look at your own account and OpenAI's site; we do not quote them here.

How subscription limits for Codex work

According to OpenAI's documentation, Codex usage is part of ChatGPT plans and consumes a shared allowance. Key points:

• The number of messages depends on the model, the size and complexity of the task, and where it runs, locally or in the cloud. Cloud tasks can use more than local messages.
• Two similar-looking tasks can consume different amounts. Model, context, reasoning, tool calls, retrieval and caching all affect usage, so prompt length alone is a poor guide.
• For some plans the allowance is counted in time windows (the documentation mentions five-hour windows), and weekly restrictions may also apply. Exact values depend on the plan and may change.
• The "messages per window" estimates OpenAI publishes are described as approximate, not fixed limits.
• If the limit runs out mid-turn, the agent may, per the documentation, finish the current turn within fair use.

A practical conclusion follows: heavy sessions (a large repository, a long history, cloud tasks) use up the allowance much faster than short questions.

Can the limit be reset

No. The limit resets itself on your plan's schedule, and the usage panel shows when the next reset is. The documentation has no way to "reset" it on demand, and a free reset does not exist. Services that promise one do not deserve trust: most often it is a path to losing an account or money. We do not describe such schemes and do not advise looking for them.

Where to see what is left:

• the Codex usage panel in your account (the usage settings page at chatgpt.com/codex);
• the /status command inside a Codex CLI session: it shows the remaining limits in the active session.

Legitimate ways to continue named in OpenAI's documentation: buy credits (for plans where this is provided), change the plan, or run extra local sessions with an API key billed at standard API rates. See the service's site for the terms and prices of these options.

Continuing on your own key

The console Codex can work with any provider that accepts the Responses API. For Zukko this means paying for tokens from a single dollar balance, with no subscription and no time windows. You set the limits yourself.

Add the provider to ~/.codex/config.toml. Create the fetch_... key in the dashboard↗ and pass it through an environment variable:

[model_providers.zukko]
name = "zukko"
base_url = "https://api.zukko.pro/v1"
env_key = "ZUKKO_API_KEY"
wire_api = "responses"

To keep your usual subscription sign-in, move the switching into a profile. Per the documentation, with codex --profile name Codex loads ~/.codex/config.toml and then overlays ~/.codex/name.config.toml with ordinary top-level keys. File ~/.codex/zukko.config.toml:

model = "gpt-6.1-sol"
model_provider = "zukko"
model_reasoning_effort = "medium"

Launch on your own key:

export ZUKKO_API_KEY=fetch_your_key
codex --profile zukko

A plain codex without a profile keeps working with your sign-in and subscription. If the profile is not picked up, drop it and put model and model_provider straight into the main config.toml, checking on your version. According to the Zukko documentation, GPT and Claude models work in Codex and GLM models do not. Setup details are in the Codex documentation↗, and for connecting without a subscription read the article Codex without a ChatGPT subscription↗.

A dollar budget instead of a mysterious limit

On your own key spending is transparent: it is made of tokens and the model price. The formula: fresh input × input price + cache read × read price + output × output price. For GPT models cache writes are not billed separately, so there is no "cache write" line in the calculation.

An example: a session of 30 requests with 50,000 input and 1,000 output tokens each. Input 30 × 50,000 = 1,500,000, output 30 × 1,000 = 30,000. Prices as of October 18, 2026 per 1M tokens.

• gpt-6.1-sol (input $0.27, output $1.32), no cache: 1.5 × 0.27 + 0.03 × 1.32 = 0.405 + 0.0396 = $0.4446.
• The same if 80% of input is read from cache (an assumption for the example, the real share depends on the session): fresh input 0.3 × 0.27 = 0.081, cache read 1.2 × 0.013 = 0.0156, output 0.0396. Total $0.1362.
• claude-sonnet-5-5 (input $0.4, output $2), no cache: 1.5 × 0.4 + 0.03 × 2 = 0.6 + 0.06 = $0.66.

If there is one such session a day and 20 working days, then at $0.4446 per session that is 20 × 0.4446 = $8.892 a month. This illustrates orders of magnitude, not a promise: your sessions may be lighter or heavier. See real spending in the dashboard under "Usage" and through the usage API. The list of models and prices: GET https://api.zukko.pro/v1/models.

How not to hit the ceiling again

A dollar limit on the key protects against surprises. A key can have a daily and monthly limit and a list of allowed models. When the limit is used up, a request gets 429 with the code key_daily_limit_reached or key_monthly_limit_reached, and if the balance ran out you get 402. These are clear signals rather than a silent block. How to set the values is described in the article How to set a daily and monthly limit on a key↗.

Ways to spend less regardless of plan:

• start a new session for each new topic: a long history is paid for at every step;
• keep AGENTS.md short and precise, it enters the context;
• use a cheaper model for routine work and keep a strong one for hard tasks (compare quality on your own tasks, not by price);
• trim command output and logs that land in the context;
• ask for a plan first, then for execution.

More ways to cut spending are collected in the article How to cut token spending in practice↗.

Top-up

The balance is a single dollar balance, topped up from $1. Available: SBP (the payment page adds an 8.5% fee on top, and the balance is credited with the dollar amount you set), cryptocurrency and MasterCard cards. There is no subscription; you pay for the tokens you use.

FAQ

How do I reset the Codex limit?

You cannot: the limit refreshes on the plan's schedule, and the current remainder and reset time are visible in the usage panel and via /status. If you need to continue now, use the options from OpenAI's documentation or your own key billed by tokens.

Is there a free Codex limit reset?

No. Do not trust services and instructions that promise a free limit reset: that is a risk for your account and money.

How do I see Codex usage?

On a subscription: the Codex usage panel and /status in the CLI. On your own Zukko key: the "Usage" section in the dashboard and the accounting API.

Can I keep a subscription and my own key at the same time?

Yes, if you move the Zukko provider into a separate profile and run codex --profile zukko only when needed. A normal launch stays on your sign-in.

If you want to try this, create a key in the Zukko dashboard, set a daily limit on it and run the profile on a small task.