Quick answer
Cloudflare now lets prepaid AI Gateway credits pay for Workers AI inference when the gateway’s Workers AI billing setting is changed to Unified billing. The same credit balance can cover Workers AI and supported third-party providers.
At this page’s source check, selected frontier Workers AI models billed with gateway credits had 50 requests per minute per account, per model, compared with 20 requests per minute through standard Workers AI billing. Limits, eligible models, pricing, and plan rules can change; always recheck the live documentation.
Billing decision table
| Question | Evidence required |
|---|---|
| Which balance pays? | Gateway billing setting and request-level cost attribution |
| Which model is eligible? | Current Workers AI limits and model catalog |
| Which limit applies? | Account, model, billing path, and approved increase |
| Who owns spend? | Application, environment, user or agent metadata, and budget owner |
| What happens at a limit? | Error handling, backoff, queueing, degradation, and alert path |
| Can a retry double cost? | Idempotency, request correlation, and reconciliation evidence |
Do not infer rate limits for every Workers AI model from the frontier-model example. Do not promise a higher limit until the specific model, account, billing path, and any approved exception are verified.
Control credits like production spend
Set budgets and alerts by application and environment. Where supported, use gateway spend limits and identity or custom metadata for attribution. Restrict who can change billing mode, load credits, edit gateways, change models, or raise limits.
Monitor request volume, tokens, cache hits, retries, errors, latency, and cost together. A low request count can still be expensive with long contexts or costly models; a retry storm can exhaust both rate and spend capacity.
Protect credentials and investigate sudden spend changes as a possible compromised key, runaway agent, cache miss, loop, or routing error. Test depleted balance, limit exhaustion, provider failure, stale pricing, and loss of billing attribution.
Reconcile before rollout
Run a bounded test and confirm each Workers AI request deducts from the intended balance. Compare dashboard and application telemetry, including model, tokens, cost, gateway, user or agent, and correlation ID. Keep standard billing and rollback behavior documented.
Use the control-plane guide for architecture and the readiness checker before enabling unified billing broadly.
Frequently asked questions
Can AI Gateway credits pay for Workers AI?
Yes. Cloudflare says prepaid AI Gateway credits can pay for Workers AI requests when the gateway’s Workers AI billing setting is set to Unified billing.
What frontier-model rate limits did Cloudflare document?
At the August 12 source check, Cloudflare documented 50 requests per minute per account per frontier model with AI Gateway credits versus 20 through standard Workers AI billing.
Does a prepaid balance replace spend controls?
No. Define budgets, attribution, alerts, rate and spend limits, retry policy, credential controls, and reconciliation for each application and environment.
Official sources
Source check: August 12, 2026. Recheck eligible models, rate limits, pricing, billing mode, spend controls, geography, and service terms.