AI coding · fixes · 2026-10-24

Codex “Selected model is at capacity”: what to do

As of 24 October 2026. This is a fast-moving problem. OpenAI may change things at any time, so check the linked issues before you follow any step.

If you want the next step, read Codex credits explained: what they cost, earned limit resets and how to cap auto reload.

If Codex stops a turn with this message, you're not alone:

Selected model is at capacity. Please try a different model.

That's the exact text quoted in openai/codex #55740 (opened 21 Oct 2026), where the reporter says it happens on all models and across Web, the desktop app, and the CLI, with quota left. Many similar issues have been opened since. OpenAI has not published a cause as of 24 October 2026.

Quick fix: what to try right now

Check status.openai.com. When we checked on 24 October 2026, it read "All Systems Operational", so a green page doesn't rule this error out.
Try a different model, as the message itself suggests. Reporters in #55740 say switching didn't help them, so treat this as a quick test, not a sure fix.
Wait and retry. One reporter in #55740 says failures came and went, not on every request.
Resume, don't restart. If a long task stopped, pick it back up with codex resume (see the Codex CLI reference) so you keep its context.

What “at capacity” means (and why it isn’t your usage limit)

It shows up on many models and plans

In #55740, a Pro user reports the error on every model they tried. A commenter on the same issue logged about 73% of their turns failing on 21 October, with weekly usage far from used up (#55740 comments). Some reporters suspect an account flag, others a capacity problem. Nobody in the thread has confirmed either, so we don't guess.

At capacity vs usage limit vs 429

At capacity (this post): Codex can't serve the request right now, even when you have quota left.
Usage limit reached: you've used your plan's allowance. See Codex credits, costs, and usage limit resets and ChatGPT Pro and Codex usage limits.
429 in your own app: your code hit the API's rate limit. See the 429 rate limit guide.

Make Codex retry on its own

The Codex config reference lists two retry settings per model provider: model_providers.<id>.request_max_retries ("Retry count for HTTP requests to the provider (default: 4)") and model_providers.<id>.stream_max_retries ("Retry count for SSE streaming interruptions (default: 5)"). Raising them may help with short blips. A commenter in #55740 reports Codex already retries and then still ends the turn with the capacity error, so don't count on it.

Keep long and unattended runs from dying

For long or codex exec runs, keep each step small and committed, so a failed turn costs you one step, not the whole job. When a session stops, resume it with codex resume rather than starting fresh.

Fallback when it won’t clear

Write a short handoff note (goal, what's done, what's next, files touched) so you can continue in another tool without losing context.
If you also use another agent, such as Claude Code, switch to it for now. Watch its own limits too: see Claude weekly limit reached.
If the error looks tied to your account, as some reporters suspect, contact OpenAI support and add your details to the GitHub issue.

FAQ

Does "at capacity" use up my Codex quota? We found no official answer. Reporters in #55740 see it with plenty of quota left.

Why does every model say at capacity? OpenAI hasn't said, as of 24 October 2026. Reporters suggest either capacity or an account flag, unconfirmed.

Is Codex down right now? Check status.openai.com. It showed no incident when we checked, while users were still reporting the error.

Can Codex retry automatically? Yes, there are retry settings in the config reference, but users report the error can still come through.

How is this different from "usage limit reached"? A usage limit means your allowance is used up. "At capacity" is reported even when it isn't.

Disclosure: this is our own paid product.

Codex Capacity Failover Kit · unattended Codex CLI runs

Running Codex unattended? Keep the run going when it hits capacity

This kit doesn't fix or prevent capacity errors, and as of October 2026 OpenAI hasn't said what causes them. For unattended codex exec runs, its runner is designed to keep your run going when Codex stops with "Selected model is at capacity" or an overload error. It waits with backoff and resumes the same session. Then, only if you turn them on, it moves to an API-key login, an OpenAI-compatible provider, or Claude Code, and alerts you when it fails over, finishes, or gives up. Every step is off by default, and paid services bill your own accounts. If you turn on a failover step, your prompt, recent context and files the agent reads go to the service you chose, under your own account and that service's terms. A step can still fail if that service is busy too. It drives the Codex CLI, not the desktop app.

$12 (plus VAT where applicable)

See what's in the Codex Capacity Failover Kit

OpenAI and Codex are trademarks of OpenAI. Anthropic and Claude Code are trademarks of Anthropic. Other names belong to their owners. Names are used only to identify products and compatibility. Stack & Ship is not affiliated with or endorsed by any of them.

Sources

Checked 24 Oct 2026: github.com/openai/codex issue #55740 and its comments; developers.openai.com/codex/config-reference; developers.openai.com/codex/cli/reference; status.openai.com.

OpenAI, ChatGPT, and Codex are trademarks of OpenAI. Claude Code is a trademark of Anthropic. GitHub is a trademark of GitHub, Inc. Git is a trademark of Software Freedom Conservancy. Names are used only to identify products and compatibility. Stack & Ship is not affiliated with or endorsed by any of them.