FAQ: plans, billing, and limits
Answers about production on Free, separate Workers plans, request and cache billing, traffic costs, and budget gates.
Edited and verified by Orange Book Editorial Team ·
This is not a permanent price promise
Limits and prices change. This page gives a verification method; use the official Pricing and Limits pages plus your account bill on the day you implement a decision.
Can the Free plan run production?
It can run a small application that tolerates its request, CPU, bundle, storage, and failure boundaries, but “can deploy” does not establish an SLA. Model peak traffic, bots, retries, and degraded behavior, and define a paid or migration path before approaching a limit.
Are a zone's Free, Pro, or Business plan and Workers Paid the same plan?
No. Workers Paid is currently a separate account-level developer-platform plan; buying Pro or Business for a zone does not grant it automatically. Separate zone, Workers, R2, D1, Images, Argo, and other product charges in the cost model.
What happens when Workers Free exceeds its daily request limit?
The current official limit is 100,000 requests per day, resetting at 00:00 UTC. A Route can fail open and bypass the Worker or fail closed with 1027. A security-critical Worker normally must not bypass silently, so choose and test this behavior explicitly.
Does a cache hit still count as a Worker request?
Yes. Workers runs before Cloudflare Cache, so the request still invokes and counts against the Worker. A hit can reduce origin and CPU work but does not erase request count. Directly served static assets follow a different path; verify the actual route in Metrics.
Does network waiting count as Worker CPU?
CPU time measures active execution, not waiting on fetch(), KV, or database responses. Requests, subrequests, and downstream storage operations can still have their own limits or charges. Model request count, CPU, and each binding's operations separately.
Does “no R2 egress fees” mean R2 is free?
No. Internet egress is not charged, but storage, Class A/Class B operations, and storage-class behavior still cost money. Listing many small objects, frequent writes/deletes, and lifecycle actions can matter; do not estimate only from stored gigabytes.
Should current price numbers be embedded in architecture decisions?
Keep a dated budget snapshot and calculation inputs, not a “free forever” claim. Record dependencies, alternatives, acceptable unit cost, and upgrade triggers. Reopen the official Pricing and Limits pages before each release.
How can I prevent an accidental bill?
Set separate limits and alerts for requests, CPU, storage operations, AI inference, and third-party APIs. Add a Worker CPU ceiling, application quotas, idempotency and queues for expensive actions, and per-tenant usage records. Use the plans, limits, and cost checklist.
Primary sources
Did this page help you complete your goal?
Beta feedback is generated in this browser and is never uploaded automatically.