For agents, developer tools and automations

Automatic Fallback for your APIs.

If the ordinary service slows down or stalls, we can handle the move to an alternative.

With Fallback, Uncensured AI can change provider when the ordinary service has problems, preserving your chosen model and variant. It helps reduce the retries and changes you would otherwise manage in your integration.

View Fallback prices ↘

Standard while everything works.

Standard

Standard prices still apply when the ordinary service works, even with Fallback enabled.

View Standard prices ↗

Fallback

If we detect a problem and use an alternative provider, that response uses Fallback prices. Enabling the option alone does not change request prices.

Without Fallback, your client or SDK manages retries and alternatives.

Fallback prices

US dollars per million tokens. Applies only to responses that actually use the alternative.

Fallback prices · USD / 1M tokens
ModelInputCache readOutputCache write
DeepSeek V4.1 Flash AbliteratedFallback availabledeepseek-v4.1-flash-abliterated$0.54$0.10$2.24$0.54
GLM 5.3 AbliteratedFallback currently unavailableglm-5.3-abliterated$1.79$0.29$3.58—
GLM 5.3 Flash AbliteratedFallback availableglm-5.3-flash-abliterated$0.54$0.22$1.79—

Version 2026-11-01-fallback1 · A dash means the tariff has no separate entry for that item.

Fallback applies only to models marked as available. Other models stay on Standard; without a compatible alternative, a request may end with the usual error. The tariff also lists reference prices for models that are currently unavailable.

How we choose the alternative.

Selection considers errors, waiting time and observed generation speed in tokens per second. When there is not enough data to compare speed, we use availability and configured priority.

These signals help choose a route; they do not guarantee a minimum speed or uninterrupted service. If no alternative is available, a request may still end with the usual error.

Before you enable it.

Do I pay more when it is not needed?

No. The setting authorizes a possible intervention, but Fallback prices apply only when a problem leads to actual use of the alternative. A failed attempt does not create a second customer charge.

Can it change my chosen model?

No. The model and variant stay the same. We do not switch from Flash to Pro, from Abliterated to Uncensored or to another family. If there is no compatible alternative, the request may end with an error.

What if a response has already started?

A switch can happen before the response starts being sent to your client. A response already in progress is not assembled from two providers: an interruption follows the usual error handling and partial usage billing.

Can I turn it off at any time?

Yes. The change applies to subsequent requests across your account. A call already in progress keeps the consent and prices captured at its start.

Uncensured AI

Loading...