Cloudflare Workers Pricing Calculator for a SaaS

How Workers bills a request

The Paid plan charges a monthly minimum, then two meters: requests that reach your Worker, and the CPU milliseconds your code uses. What your Worker calls in turn, such as fetch(), KV, R2 or D1, is not billed as more Worker requests; those products bill on their own meters, and time spent waiting on them is not CPU time. A call to another Worker through a service binding is not a second request either, though its CPU time adds to the first.

Workers Caching moves cost between lines. With it on, static asset requests and service binding calls are billed at the request rate too, while every cache hit skips your code, so CPU, D1 and KV fall. Whether it saves money depends on the hit share: an app that serves mostly static assets can pay more with it on.

The Free plan counts by the day

The calculator compares a thirtieth of each monthly figure with the Free plan's daily limit, which assumes even traffic. Real traffic is lumpy, so leave room for a launch day or a busy crawler.

Past a Free limit, Cloudflare refuses the extra work instead of billing it. A Worker over its daily requests returns error 1027, or is skipped if its route fails open. KV, D1, Hyperdrive and Workers AI return errors until the reset. CPU is limited per request rather than per day, and a Worker that keeps going over it is stopped with error 1102 at any hour.

D1 bills rows scanned

D1 counts every row a query reads on the way to its answer, not the rows it returns, so a filter on a column without an index reads the whole table. An index cuts the rows read and adds one written row to each write that sets the indexed column. Take the rows per request from meta.rows_read on your real queries rather than guessing.

The D1 pricing calculator goes further: daily figures, several databases, and how many rows fit in one INSERT before D1 refuses it.

Postgres through Hyperdrive

Hyperdrive puts a connection pool and a query cache between a Worker and a Postgres or MySQL database. It counts every query, cached or not, against the Free plan's daily limit, so a cache that spares your database does not spare the limit. The database itself bills through its host, which this calculator leaves out.

Workers AI in neurons

Workers AI measures GPU work in neurons and prices each model's tokens in neurons per million. The same reply costs far more on a large model than on a small one, so compare the model rates before you choose one. Embedding models such as bge-m3 have an input rate only. Both plans share the same daily allocation, and only Paid can go past it.

How much CPU a request takes

Cloudflare's limits page puts the average Worker at about 2.2 ms of CPU a request, and workloads that render on the server or handle authentication at 10 to 20 ms. A Nuxt app that renders its pages on the server belongs to that second group, so check its figure against the Free plan's limit per request before you count on Free. Workers Logs records the CPU time of every invocation, and so do Tail Workers and Logpush.