Hit the Codex Usage Limit? Check Reset Times and Keep Working (2026)

Last verified: September 18, 2026. Codex limits, model availability, credit rates, and commands can change; check the linked official pages before publishing.

When Codex stops with “You’ve hit your usage limit,” “You are out of Codex and Work usage,” or “You have reached your Codex usage limits for security reviews. Please try again later,” the fastest fix is not to keep retrying. First identify which limit you reached, then choose the correct recovery path.

This guide explains how to check the current reset time, how the five-hour allowance and possible weekly limits work, why Codex and other agentic features can consume the same pool, how context limits differ from usage limits, and how to continue an urgent local CLI task with a separately billed API provider.

TL;DR

• In an active Codex CLI session, enter /status. For account-level details, open the Codex Usage Dashboard.
• Do not treat every failure as the same problem. A ChatGPT plan limit, an API 429, a security-review failure, and context_length_exceeded have different causes.
• OpenAI currently publishes estimated local messages per five-hour period, not a guaranteed fixed message count. Local and cloud tasks share an allowance, and weekly limits may also apply.
• The official recovery choices are to wait for the displayed reset, reduce usage, use purchased credits where available, change plan, or run additional local work with an API key. An API provider does not reset your subscription allowance; it uses a separate billing path.

How to check your Codex limit in 5 seconds

In the running Codex CLI session, enter:

/status

/status shows the current chat ID, context usage, and rate-limit information. Then open Settings → Usage or the Codex Usage Dashboard to see account allowance, credits, and the reset time shown for your account.

Some third-party posts recommend /usage. As of September 18, 2026, the official Codex slash-command reference lists /status, but does not list /usage. If /usage is unrecognized, that is expected; use /status plus the dashboard.

Codex error decoder

Message you see	What it usually means	What to do next
You've hit your usage limit or 5-hour limit reached	Your included ChatGPT/Codex allowance for the current period is exhausted.	Check /status and the dashboard. Wait for the displayed reset, use credits if available, choose a lighter model, or move urgent local work to API billing.
You are out of Codex and Work usage	Codex and supported agentic products are drawing from the same included usage or credit pool.	Check which product consumed the pool, then follow the reset or credit options shown in Settings.
You have reached your Codex usage limits for security reviews. Please try again later.	The failure is on the GitHub-hosted Codex security-review path, not necessarily your local CLI chat. It may be a review-specific quota or a product issue.	Record the repository, PR, timestamp, and failed review. Check the official Codex GitHub issue and contact support if the dashboard does not explain it. Pause repeatedly triggered automatic reviews while diagnosing, but do not assume that disabling automation restores quota.
429 Too Many Requests or rate_limit_exceeded	An API organization, project, model, TPM/RPM, concurrency, or provider-side rate limit was reached.	Inspect the API provider’s usage and limits, reduce parallel requests, add backoff, or request a higher limit. Waiting for a ChatGPT subscription reset may not help.
insufficient_quota	The API project has no usable balance, spend permission, or billing capacity.	Check API billing, project budget, key scope, and provider balance.
context_length_exceeded	The current request contains more history, files, or tool output than the selected model/provider accepts.	Use /compact, start a fresh chat with a handoff note, remove large attachments, or choose a model with a larger supported context. This is not the same as a five-hour usage limit.

Which limit did you actually hit?

1. ChatGPT plan or agentic usage limit

This is the common case behind codex hit usage limit, codex usage limit, and you've hit your usage limit codex searches. The limit is tied to your plan and current usage. Task complexity, selected model, reasoning effort, context size, local versus cloud execution, and other agentic products can all change how quickly the allowance is consumed.

2. API rate or billing limit

API usage is billed and limited separately from the included ChatGPT plan allowance. API credits do not increase your ChatGPT/Codex subscription allowance, and ChatGPT credits do not become API balance. A 429 or insufficient_quota therefore requires checking the API account or provider that served the request.

3. Context-window limit

A context limit is about the size of one request or conversation, not the amount of work allowed during five hours. There is no single universal 272k value that applies to every Codex model, client, and custom provider. Use /status to watch context usage, then compact or hand off before the session becomes unwieldy.

4. GitHub security-review limit

The exact security-review message has appeared in the official openai/codex issue tracker. Public evidence confirms the failure text, but it does not establish one universal root cause or a guaranteed workaround. Treat it as a distinct GitHub review incident: collect evidence, check service/account status, avoid repeated automatic triggers, and escalate with the PR details.

How the five-hour period, weekly limits, and shared pool work

Many pages call this a “rolling five-hour window.” The current official pricing page publishes estimated local messages per five-hour period and directs users to the dashboard for their actual limit and reset. In practice, do not assume a midnight reset, a fixed clock-hour reset, or a universal date. Follow the reset timestamp shown for your account.

Three details cause most confusion:

1. The message count is an estimate, not a promise. A short edit and a long agent run with tools, large context, and high reasoning do not consume the same amount.
2. Local and cloud tasks share the allowance. Moving from the CLI to a cloud task does not necessarily create a fresh pool.
3. Weekly limits may apply. A five-hour period can reset while a weekly cap still prevents more included usage.

The exact search phrase “Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents draw from the same agentic usage and credit pool” captures the core idea, although the list of supported products changes by plan and availability. Current OpenAI documentation also mentions agentic experiences in Word and PowerPoint where available. Always use the product list shown in your own Settings page.

Current five-hour estimates by model

The following ranges are the official estimates for local messages per five-hour period, verified on September 18, 2026. They are not guaranteed message quotas.

Model	Plus	Pro 5×	Pro 20×	Standard Business
GPT-6 Astra	5–45	25–225	100–900	5–45
GPT-5.6 Sol	10–100	50–500	200–2,000	10–100
GPT-5.6 Terra	25–200	125–1,000	500–4,000	25–200
GPT-5.6 Luna	250–2,000	1,250–10,000	5,000–40,000	250–2,000

A range this wide is normal: repository size, prompt length, tool calls, cached context, reasoning level, and fast mode all affect consumption. Use the table for planning, not for predicting an exact reset after a specific number of prompts.

Purchased-credit rates by model

When purchased credits are available for your plan, OpenAI currently measures model usage in credits per one million tokens:

Model	Input	Cached input	Output
GPT-6 Astra	250	25	1,250
GPT-5.6 Sol	100	10	500
GPT-5.6 Terra	50	5	300
GPT-5.6 Luna	5	0.5	30

A typical GPT-5.6 Sol Codex task is currently estimated at roughly 5–30 credits, but real consumption depends on the task. Check the live rate card before making budget decisions.

Option A: stay on the official included-usage path

Use this route when the task is not urgent or when you need cloud-only features:

1. Run /status and open the Usage Dashboard.
2. Note whether the blocker is the five-hour period, a weekly limit, or zero credits.
3. Wait for the displayed reset, buy credits where your plan supports them, or upgrade if the economics make sense.
4. Reduce consumption by choosing Terra or Luna for routine work, lowering reasoning effort, narrowing the request, and starting a fresh session when old context is no longer useful.
5. If the displayed usage looks wrong, capture the exact error, timestamp, account, client version, and task type before contacting support.

Do not repeatedly resubmit the same large task. Retries can consume more capacity without changing the underlying limit.

Option B: continue an urgent local task with a separate API budget

For a local CLI task that cannot wait for the subscription reset, Codex supports custom model providers. A provider such as BetterToken can route the local Codex CLI through an OpenAI-compatible endpoint and bill by token.

This is a separate budget, not a reset or loophole. It does not add subscription allowance, remove API rate limits, unlock plan entitlements, or provide cloud/GitHub features that require the official hosted environment.

Step 1: preserve a clean handoff

Before switching providers or starting a fresh session, save a short handoff note outside the chat:

Goal:
Current state:
Files already changed:
Commands already run:
Known failures:
Next smallest step:
Do not change:

For an unfinished edit, also save the working tree or patch by your normal development process. The purpose is to make the next session reconstructable without carrying the entire context.

Step 2: find CODEX_HOME and check the CLI version

Codex uses ~/.codex by default unless CODEX_HOME is set.

macOS or Linux:

export CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
mkdir -p "$CODEX_HOME"
echo "$CODEX_HOME"
codex --version

PowerShell:

$CodexHome = if ($env:CODEX_HOME) { $env:CODEX_HOME } else { Join-Path $HOME ".codex" }
New-Item -ItemType Directory -Force -Path $CodexHome | Out-Null
$CodexHome
codex --version

Named profile files such as bt.config.toml require Codex CLI 0.134.0 or later. Update the CLI first if your version is older.

Step 3: create an isolated BetterToken profile

Create $CODEX_HOME/bt.config.toml with this content:

model = "gpt-6-astra"
model_provider = "bettertoken"

[model_providers.bettertoken]
name = "BetterToken"
base_url = "https://www.bettertoken.ai/v1"
env_key = "BETTERTOKEN_API_KEY"
wire_api = "responses"
requires_openai_auth = false
request_max_retries = 4
stream_max_retries = 8
stream_idle_timeout_ms = 300000
supports_websockets = false

This keeps the provider settings out of the main config.toml. Do not put the API key in this file or commit it to a repository.

BetterToken’s current setup page may also present an auth.json-based flow. Do not mix the environment-variable profile above with a different authentication pattern. Use one complete method, and treat the current setup page as the source of truth if authentication requirements change.

Step 4: run a minimal read-only test

macOS or Linux:

export BETTERTOKEN_API_KEY="YOUR_API_KEY"
codex exec --profile bt "Read README.md and summarize its purpose. Do not modify any files."

PowerShell:

$env:BETTERTOKEN_API_KEY="YOUR_API_KEY"
codex exec --profile bt "Read README.md and summarize its purpose. Do not modify any files."

Replace YOUR_API_KEY locally. Never paste a real key into a ticket, screenshot, chat, or repository. A read-only prompt is safer than resuming the full task immediately because it verifies authentication, endpoint compatibility, model access, and response streaming with minimal cost and risk.

Step 5: verify billing before resuming

Confirm that the test appears in the provider’s usage or balance page. Then start the real task with the handoff note and a narrowly scoped first action. If the test fails, read the exact error before changing multiple settings at once:

• 401 or 403: key, account, authentication mode, or model permission.
• 404: base URL, route, or model ID.
• 429: provider rate limit or balance policy.
• Streaming timeout: network path or provider timeout; do not immediately increase every retry value.

How to use less Codex allowance next time

• Use GPT-5.6 Terra or Luna for file discovery, formatting, small edits, and routine checks; reserve Astra or Sol for work that truly needs deeper reasoning.
• Ask for one verifiable change at a time. Broad “inspect everything and fix everything” tasks usually consume more context and tool calls.
• Start a new session with a handoff note when the old conversation is mostly historical baggage.
• Run /compact before context becomes the bottleneck, but verify that critical constraints remain in the compacted summary.
• Disable unused MCP servers and avoid attaching large generated logs unless they are needed.
• Stop after the smallest relevant verification passes instead of running unrelated full suites.
• Use the Usage Dashboard as the source of truth rather than estimating from the number of messages sent.

Frequently asked questions

When does the Codex usage limit reset?

There is no universal reset clock published for every account. Open /status and the Usage Dashboard and follow the timestamp displayed there. Do not infer your reset from midnight, another user’s screenshot, or an article publication date.

Does “August 14, 2026” mean Codex resets globally on that date?

No. A date such as August 14, 2026 may be an article update, an account-specific timestamp, or the date of a report. It is not a universal Codex reset date. Your dashboard is authoritative for your account.

Why are my Codex limits different from another user’s?

Plan tier, model, task complexity, context size, reasoning effort, fast mode, local/cloud mix, shared use by other agentic products, purchased credits, and temporary product changes can all affect effective capacity. A post saying “we’ve investigated a few messages about Codex usage limits being different” may describe one moment in time; compare it with the current official pricing and your dashboard.

Can I check the limit with /usage?

The current official CLI reference documents /status, not /usage. Use /status for the active session and the web Usage Dashboard for account-level allowance, credits, and reset information.

Is a context-window error the same as a usage limit?

No. A usage limit controls how much agentic work is included or allowed during a period. A context limit controls how much material one request/session can carry. Compact, remove files, or start a fresh session for context errors; waiting for a five-hour reset does not shrink the conversation.

Does an API key bypass Codex limits?

It does not bypass or reset the ChatGPT plan limit. It sends local work through a separately billed API route with its own balance, model access, and rate limits. This can keep an urgent local task moving, but it is not free and may not support hosted features.

Continue without losing the task state

The safest sequence is simple: check /status → identify the limit → save a handoff → choose the official reset/credit path or a separate API budget → run a read-only test → resume with a narrow step.

BetterToken is useful when the blocker is an exhausted included allowance and the work must continue locally. It is not the answer to every error. For a GitHub security-review incident, an API 429, or an oversized context, fix the specific cause first.

Official references

• Using Codex with your ChatGPT plan
• Codex pricing and five-hour estimates
• Codex slash commands
• Advanced Codex configuration and custom providers
• ChatGPT agentic credit rate card
• Official Codex issue: security-review usage-limit message
• BetterToken Codex setup