Codex ‘Selected model is at capacity’: Meaning and How to Resume
In Codex, “Selected model is at capacity. Please try a different model.” means that the selected model currently lacks spare capacity to handle another request. Wait and retry, or choose another available model, but first check what was already completed and continue only the unfinished work.
What does “Selected model is at capacity” mean?
Selected model is at capacity. Please try a different model.
The message says that the selected model cannot currently accommodate more work and suggests trying another model. Here, capacity refers to the model service's ability to accept and process requests. The wording does not indicate that your computer has run out of disk space.
An agent may call the model again during a task to decide what to do next or generate a response. If that request cannot be processed, the task can appear to stop. This does not establish that earlier file saves, commands or changes in external services were canceled. Check when the error occurred and the actual outcome of each action separately.
OpenAI's official API error guide distinguishes temporary model overload from usage and request limits. An application programming interface (API) is how software communicates with a service. However, this Codex message alone does not reveal the actual HTTP status or internal error code. The API guide helps explain the distinction; it does not prove a fixed one-to-one mapping between this app message and a particular code.
How is it different from exhausting a personal usage allowance?
The model's processing availability and your account's remaining allowance are separate conditions. Temporary model-capacity problems can occur while allowance remains. Conversely, this message alone does not prove that your account has plenty of usage left, so check any separate usage-limit notice as well.
| Issue | What to check | Response |
|---|---|---|
| Model capacity | Temporary ability of the selected model to accept work | Wait and retry, or consider another available model |
| Account usage or request limit | Usage notices, reset information and request-limit messages | Follow the stated conditions and check the relevant limit |
| Conversation or context limit | A separate notice that the input exceeds what the model can process at once | Summarize necessary context and reduce unnecessary input |
| Connection, authentication or execution environment | Specific connection, sign-in, permission or tool errors | Investigate the reported issue and address its cause |
Context is the conversation and material available for the model's current reasoning. A capacity message during a long task is not, by itself, evidence of a context overflow. OpenAI's agent errors and recovery documentation treats overload, context limits, usage limits and connection failures separately. If disappearing or queued requests are the main symptom, distinguish them from the behavior discussed in Codex VS Code extension request and queue problems.
A practical sequence for resuming a stopped task
- Record the error and latest progress. Keep the message, selected model, time and last confirmed completed item. If a command or upload was running, check its outcome before starting the same operation again.
- Wait before retrying. Space attempts out instead of repeatedly sending the same request. Increase the interval if failures continue. The message does not establish a guaranteed recovery time.
- Consider another model if the issue persists. If a model selector is available in the current conversation, choose another model offered to your account. Check that it supports the tools and capabilities your remaining task needs.
- Reconcile existing results and request only unfinished work. Ask the agent to inspect saved files or the external service before continuing, rather than telling it to recreate and rerun everything.
- Verify resumption with a small remaining step. For example, ask it to inspect the last saved file before moving on. If the capacity message returns, do not assume work continues; choose whether to wait, change models or report the issue.
The official model selection guide notes that availability, tools, reasoning settings and usage limits vary by product and model version. Switching to a particular model is therefore not a guaranteed fix. For developers working directly with the API, respect Retry-After when the response provides it. This retry-delay header is not necessarily exposed to app users.
What should you check about already completed work?
An interrupted conversational response does not mean that nothing changed. A file may have been saved before the completion report failed, or an external service may have processed a request before the error appeared. Inspect each outcome instead of classifying the entire task as successful or unsuccessful from one message.
| Task example | Check before resuming | Avoid duplication by |
|---|---|---|
| Writing or editing files | Latest contents, saved location and changes | Reading the current file and changing only what remains |
| Publishing blog posts | Post ID, publication status, public URL, language and category | Finding existing posts before deciding whether to create another |
| Uploading media | Media ID and associated post | Checking previous uploads before adding more |
| Commands or external tools | Outcome, change records and current execution state | Confirming effects before repeating an uncertain operation |
Suppose a seven-language publishing task stops after three posts have been published. First check the actual posts and translation links, then handle only the other four. This is an illustrative recovery scenario, not a claim that every failure leaves the same number of completed actions. The official recovery guidance likewise advises checking file and external-tool effects before retrying.
A prompt you can use to continue
A resumption request is more useful when it includes the goal, where to inspect completed results, the remaining work and what must be preserved. In the existing conversation, you can use wording such as:
The task stopped with a model-capacity error. First inspect the current files and the results actually saved in external services. Do not repeat completed actions; continue from the unfinished work. Preserve the content, URLs and featured images of published posts outside the requested changes. At the end, distinguish completed items from anything still outstanding.
If you need a new conversation, include the task goal, relevant folder and file paths, created post or media IDs, completed and unfinished lists, and applicable instructions. Do not assume that a new model or conversation automatically knows every previous result. Supply a work record and ask it to verify the actual state.
After resuming, verify outcomes rather than only an acknowledgment that work will continue. Check current contents for a file save, or the public page and status for publication. Ask the agent to avoid creating or publishing another item until any uncertain external operation has been reconciled.
Frequently asked questions
Can this happen while I still have usage available?
Yes. Current model-service capacity and account allowance are separate conditions. The message does not reveal your remaining allowance, so check any separate usage information too.
Do I need to reinstall the app or change my paid plan?
This message alone does not establish a need to reinstall or change billing. Start with waiting, retrying and considering another available model. If a different error appears as well, investigate that specific issue.
What should I record if it keeps happening?
Record the time and time zone, model, app version, exact wording, task in progress, whether another model has the same issue, and completed results. Check the service-status guidance linked from the official error documentation and contact support if it persists. Remove passwords, authentication tokens and private material from shared screenshots or logs.
