Reader guide · Reviewed September 20, 2026
AI coding-agent limits: usage, fallback models, and access
A model appearing in an app does not prove that your account can use it. Check the allowance, the model selected after a limit, and the account's live eligibility separately.
Four different limits to check
| Mechanism | What it means | Evidence to check |
|---|---|---|
| Usage allowance | How much work an account can run before a limit applies. | The account's current usage screen and the provider's plan documentation. |
| Fallback model | A different model path used when a primary allowance is exhausted. | Which model runs, whether its quota is separate, and how the original selection is restored. |
| Rollout gate | A server-controlled eligibility check that can restrict an implemented feature. | Whether the feature is actually enabled for the account; client code alone cannot establish this. |
| Capacity restriction | A limit on access or purchases while a provider manages demand. | The dated announcement, affected plans, and whether existing subscribers retain access. |
Example: Codex Luna Reserve
Our reporting, updated September 2, describes a separately metered fallback named Luna Reserve in the Codex Windows client. The code preserves the user's original model selection and restores it after the reserve period. A server-controlled feature gate means that this is evidence of an implemented workflow, not proof of universal availability.
Read the evidence and limitations in our Codex reportMethodology and limitations
This comparison separates mechanisms described in our published reporting. It is not a benchmark, an account eligibility test, or a current price list. We distinguish observed interface or code behavior from provider statements and account-level access. Publication and update dates on the linked reports identify the evidence period.
Before choosing a plan, verify the provider's current terms and the allowance shown in your own account. Do not infer performance or access from a model name alone.