Codex usage limit reached?
Check your reset time and choose how to continue.
If Codex says “usage limit reached,” check your current allowance and reset time before paying for anything. For ChatGPT subscription usage, you can wait for the relevant limit to reset, buy extra credits if your account is eligible, or consider a larger plan. For supported local work, you can also use a separately billed API key.
The right choice depends on what stopped you. A ChatGPT usage limit, an API rate limit, and an exhausted API balance need different fixes. Adding a key does not refill your subscription allowance or grant Codex cloud access.
Checked September 27, 2026. This guide uses current OpenAI documentation and OmniaKey's published catalog. We did not exhaust a paid account or run a paid switching test. The cost example is a calculation, and your account's dashboard remains the source for its limits and reset times.
First, identify which limit you reached
Start with the sign-in method and the complete error message. A subscription warning belongs to the ChatGPT account you signed into. API errors belong to the organization, project, or gateway serving the request.
| What you see | What to check | Useful next step |
|---|---|---|
| Usage-limit message with ChatGPT sign-in | Current allowance and the displayed reset time | Wait, use eligible extra credits, or compare plans |
| API 429 saying requests are too frequent | Error code, request/token rate, Retry-After | Reduce concurrency and follow the server's retry delay |
| API balance or spend-limit error | Balance and limits at the selected provider | Restore the relevant balance or review the limit with its owner |
| 401, model-not-found, or access error | Key, provider, exact model ID, permissions | Correct the configuration or access; waiting for a subscription reset is unrelated |
HTTP 429 alone does not identify the problem. OpenAI's error guide also uses 429 for codes such as credit_balance_exhausted and project_spend_limit_exceeded; the broader type can be insufficient_quota. Repeating those requests will not restore a balance. Gateways can use different error names, so read the actual response.
For a genuine rate limit, follow Retry-After when present. Otherwise use bounded exponential backoff with jitter. OpenAI notes that unsuccessful requests can contribute to the per-minute limit, so rapid retries can make recovery harder.
How to check Codex usage and reset times
For ChatGPT-backed usage, open the Codex usage dashboard in the account and workspace you use for coding. In an active Codex CLI session, run:
/status
The official pricing page documents both places to inspect usage. If you need to identify the CLI's stored authentication method, use:
codex login status
A custom-provider configuration is another part of the picture: also check the selected model_provider, model, and provider URL. Stored ChatGPT login details alone do not prove that a particular request is being billed to ChatGPT. For an API route, use that provider's usage records to verify the bill.
Five-hour limits and weekly limits are different
OpenAI publishes estimated local-message ranges per five-hour period, and says weekly limits may also apply. Local messages and cloud chats share your plan's allowance. These are not fixed daily message counts: model choice, task complexity, accumulated context, reasoning and tool use all affect consumption.
Read each limit's reset time in your own dashboard. A short-window reset does not establish that a separate weekly allowance has recovered. There is no universal “every Monday at midnight” reset time established by the documentation used here.
If you search for “Codex weekly limit reset,” the most useful answer is the timestamp attached to your account's weekly limit, not a timer copied from another user's screenshot. Similarly, the same number of prompts can consume different amounts on different models.
What to do after the Codex usage limit
Choose by how urgently you need to continue and which features your work requires.
| Option | A good fit when | What to compare |
|---|---|---|
| Wait for reset | The work can pause until the displayed time | Whether the short window, weekly limit, or both are exhausted |
| Buy extra ChatGPT credits | You want to stay in the current workflow and your account is eligible | The purchase terms and consumption rates shown to your account |
| Change subscription plan | Heavy usage is recurring and the larger plan's features are useful | Total recurring cost, included usage and workspace requirements |
| Use an API for local work | You want metered model usage in a compatible local client | API prices, rate limits, feature support and a separate spending budget |
OpenAI currently says Plus and Pro users can buy additional credits after reaching their usage limit. Eligible Business, Edu and Enterprise workspaces with flexible pricing can buy workspace credits. Check the options your account actually offers before assuming a purchase path exists.
The same page says that an active turn can continue after the limit is reached, subject to fair use limits. That is not a promise of unlimited new turns. If a task is still running, save or inspect its changes before restarting the client or switching the billing path.
An occasional overage may be easiest to solve with extra credits. A higher subscription can make sense for sustained use. An API is useful when you want a separate, measurable budget for local tasks; it is not automatically the cheapest option for every workload.
Codex API key vs ChatGPT: which bill changes?
OpenAI's authentication documentation distinguishes ChatGPT sign-in for subscription access from API-key sign-in for usage-based access. The CLI and IDE extension support both for local work. Codex cloud requires ChatGPT sign-in.
Keep three payment paths separate:
- ChatGPT subscription and extra credits: governed by the eligible ChatGPT account or workspace.
- Direct OpenAI API: model usage billed to the API organization and project behind that key.
- OmniaKey custom provider: supported requests billed to the OmniaKey key and balance, at OmniaKey's current rates.
A Plus subscription does not make a separately configured API route free. Buying OmniaKey credit does not reset ChatGPT usage. An OmniaKey key belongs in its custom provider configuration, not in a field intended for a direct OpenAI key.
For a broader workflow comparison, see Claude Code vs Codex. This article focuses on the choice after a limit, rather than every feature or subscription tier.
A separate OmniaKey profile for local Codex
Use a current CLI and check the version with codex --version. In Codex CLI 0.134.0 and later, named profiles live in separate files. Create or carefully merge the following into ~/.codex/omniakey.config.toml, keeping your existing main configuration:
model = "gpt-6-sol"
model_provider = "omniakey"
[model_providers.omniakey]
name = "OmniaKey"
base_url = "https://api.omniakey.com/v1"
env_key = "OMNIAKEY_API_KEY"
wire_api = "responses"
requires_openai_auth = false
Set OMNIAKEY_API_KEY in the local shell using your OmniaKey key, then start a small new task with:
codex --profile omniakey
The current profile documentation says --profile no longer reads legacy [profiles.name] tables in config.toml from version 0.134.0 onward. If you use an older CLI, update it or follow the configuration instructions for that version.
This example changes the selected model provider without requiring you to delete ChatGPT credentials. Verify that the session selects OmniaKey and gpt-6-sol, then check the resulting call in OmniaKey's usage view before starting a large job. Workspace policies and project or command-line overrides can affect the effective configuration.
Do not combine this environment-key route with requires_openai_auth = true: OpenAI documents that this setting makes Codex ignore env_key. For the complete key setup, see the Codex CLI guide. Check the current model catalog for the exact route you intend to use.
What would an API workload cost?
Calculate the full token workload, including repeated context, instead of treating one user prompt as one API call. A coding task may read files, invoke tools, retry and send multiple model requests.
On September 27, 2026, the GPT-6 Sol page lists OmniaKey prices of $0.225 per million input tokens and $1.35 per million output tokens. For 100,000 uncached input tokens and 10,000 billable output tokens:
(100,000 / 1,000,000 × $0.225)
+ (10,000 / 1,000,000 × $1.35)
= $0.036 USD
This is a token calculation at the gateway's quoted rates, not a measured cost for a completed coding task or an OpenAI subscription price. Cache reads have their own rate; additional billable tokens, retries and separately charged tools need to be counted where applicable. Check the current quote before budgeting.
To compare API use with extra ChatGPT credits, measure equivalent work in each billing system. Do not turn a subscription's estimated message range into a fixed token allowance. Keep a modest initial budget and compare accepted changes, including the time spent correcting them.
For model selection, use the GPT-6 Sol vs Luna comparison and current prices. A cheaper model per token can still require more attempts on a difficult task.
Frequently asked questions
Does restarting Codex reset its usage limit?
Restarting does not itself replenish a server-side allowance. Check the account dashboard first. If the client appears stale after a displayed reset, check that it uses the expected account and refresh its status before changing billing settings.
Can I keep using Codex after the weekly limit?
Check whether your account offers extra credits. Separately billed API use is another option for supported local work. Neither path means that the weekly allowance itself was reset, and an API key does not add cloud entitlements.
Why did I reach the limit after only a few requests?
The number of visible messages is not the amount of work performed. Long context, tool results, reasoning and repeated agent turns consume resources. OpenAI also says speed configurations and image generation can use included limits faster. Narrow the task and supply relevant files before choosing a smaller model.
Why do I still see a limit after setting an API key?
Confirm which provider the session actually selected. Check the profile, environment variable and any higher-priority overrides. Then read the error: the API may have its own rate, balance, project or organization limit. Model availability errors belong to a different problem; the Codex compatibility guide explains how to separate those layers.
Does an API key mean unlimited Codex?
No. API use is metered and subject to provider limits, available balance, model access and applicable workspace controls. It gives local work a separate billing path; it does not promise unlimited capacity or every ChatGPT feature.