On this page
Cloudflare's free tier is generous, but it isn't one quota. It's a dozen separate counters: some per day, some per month, some per request. They don't run out in the order you'd guess from the headline numbers. This page lists the limits that matter for a typical Workers + KV + D1 project, shows what real requests actually consume, and tells you which wall your project will hit first.
The free limits that actually run out
These are the usage limits from Cloudflare's documentation as of October 2, 2026. Structural limits like script size and number of Workers are left out, because a small project never reaches them.
| Product | Free limit | Period | What happens when you exceed it |
|---|---|---|---|
| Workers requests | 100,000 | per day (00:00 UTC) | Requests fail with Error 1027 |
| Workers CPU time | 10 ms | per request | That request is terminated |
| Workers subrequests | 50 | per request | Further fetch() calls throw |
| KV reads | 100,000 | per day | Further reads fail with an error |
| KV writes | 1,000 | per day | Further writes fail with an error |
| KV deletes / lists | 1,000 each | per day | Further operations of that type fail |
| D1 rows read | 5,000,000 | per day | Queries fail until reset |
| D1 rows written | 100,000 | per day | Writes fail until reset |
| D1 storage | 500 MB per database, 5 GB total | — | Writes fail |
| Durable Objects (SQLite only) | 100,000 requests, 5M rows read, 100K rows written | per day | Calls fail until reset |
| R2 | 10 GB storage, 1M Class A ops, 10M Class B ops | per month | Billed; R2 must be enabled in the dashboard first |
Two things stand out before any testing. First, KV writes (1,000/day) are 100 times smaller than Workers requests (100,000/day). Second, D1 doesn't count queries. It counts rows, and Cloudflare's pricing page says a full table scan counts every row it scans, not the rows it returns. That second point is the one most people miss, so we measured it.
How we tested
On October 2, 2026, we deployed a test Worker on a free-plan account, bound to a new KV namespace and a new D1 database. We seeded two identical D1 tables with 10,000 rows each (an id, a slug like post-7777, a views counter and 200 bytes of body). The only difference: posts_idx has a unique index on slug and posts_noidx doesn't. We then ran common queries and read the meta.rows_read and meta.rows_written fields that D1 returns with every result. These are the numbers Cloudflare bills against your daily allowance.
We didn't drain KV's 1,000 daily writes on purpose. That limit applies to the whole account, and the account has other namespaces in use. R2 wasn't tested either: on this account the API answered "Please enable R2 through the Cloudflare Dashboard", so the R2 figures above come from the documentation only.
What one D1 query really costs
| Query (10,000-row table) | Rows returned | Rows read | Rows written | Runs per day before the free limit |
|---|---|---|---|---|
| Find a post by slug, no index | 1 | 10,000 | 0 | 500 |
| Find a post by slug, indexed | 1 | 1 | 0 | 5,000,000 |
| Find a post by primary key | 1 | 1 | 0 | 5,000,000 |
SELECT COUNT(*) | 1 | 10,000 | 0 | 500 |
Top 10 by views (column not indexed) | 10 | 20,000 | 0 | 250 |
Latest 10 by id | 10 | 10 | 0 | 500,000 |
View counter +1, lookup not indexed | 0 | 10,000 | 1 | 500 |
View counter +1, lookup indexed | 0 | 1 | 1 | 100,000 (write limit) |
| Update a row where an indexed column changes | 0 | 1 | 2 | 50,000 (write limit) |
| Insert one row into the indexed table | 0 | 0 | 2 | 50,000 (write limit) |
Seed 10,000 rows (one INSERT … SELECT) | — | 10,001 | 10,000 / 20,000 with index | — |
The last column assumes that query is the only thing your project runs. A real page does several, so the numbers add up.
Three things to take from this:
- An unindexed
WHEREreads the whole table. The lookup returned one row but was billed for 10,000. It also took 3.96 ms inside D1, compared with 0.14 ms for the indexed version. The cost grows with the table: at 50,000 rows, one unindexed lookup per page view would use up the daily rows-read allowance after 100 views. - Sorting on a column without an index costs double. "Top 10 by views" read 20,000 rows to return 10. A "popular posts" sidebar on every page would exhaust D1 after 250 page views.
- Indexes cost you on writes. Each insert into the indexed table counted 2 rows written, and seeding it cost 20,000 rows instead of 10,000. That's still cheap next to the read savings, but it halves your write ceiling to 50,000 a day.
KV: the small number that matters
KV reads are cheap and fast. In our Worker, a get() came back in 5–8 ms, and you get 100,000 a day, the same as your request budget. Writes are a different story. Each put() took 351–372 ms to return in our test, and the free plan allows only 1,000 writes a day across the whole account.
Cloudflare also documents a rate limit of one write per second to the same key. We wrote the same key four times in a row, and all four succeeded. Because each write took about 360 ms, that came to roughly three writes a second. So you won't necessarily see an error at low volume, but a hot key such as a global counter is still a bad fit for KV. KV is built for data that's read often and written rarely.
Which limit your project hits first
| What your Worker does per page view | First limit to run out | Page views per day |
|---|---|---|
| Serves a response, no storage | Workers requests | 100,000 |
| Reads one value from KV | Workers requests / KV reads (tie) | 100,000 |
| Writes one value to KV (counter, session, rate limit) | KV writes | 1,000 |
| Indexed D1 lookup + indexed view counter | D1 rows written | 100,000 |
| Unindexed D1 lookup on a 10,000-row table | D1 rows read | 500 |
| "Popular posts" sort on an unindexed column, 10,000 rows | D1 rows read | 250 |
| Any of the above, with heavy JSON or crypto work | Workers CPU (10 ms per request) | Individual requests fail, regardless of volume |
So the honest answer to "which quota runs out first" depends on two choices in your own code: whether you write to KV per request, and whether your D1 queries use an index. Get both right, and the first wall really is the 100,000 Workers requests a day.
How to stay on the free plan longer
- Move per-visit writes from KV to D1. A view counter in D1 with an indexed lookup costs 1 row written per visit, which gives you a 100,000/day ceiling instead of KV's 1,000.
- Index every column you filter or sort on. In our test, that took one lookup from 10,000 rows read to 1. Run
EXPLAIN QUERY PLANon your hot queries, and fix any that reportSCANinstead ofSEARCH … USING INDEX. - Don't run
COUNT(*)on every request. Compute the total once (from a cron Worker, for example), store it in KV, and read it from there. Reads are the cheap direction. - Read
meta.rows_readwhile you develop. Every D1 result includes it. Logging it for your top queries shows the real daily ceiling before your users find it. - Cache whole responses. A request answered from Cloudflare's cache never reaches D1 or KV.
When paying $5 makes sense
Workers Paid costs $5 a month per account. It includes 10 million requests, 1 million KV writes, 25 billion D1 rows read and 50 million D1 rows written per month. It also replaces the 10 ms CPU cap with a monthly pool of 30 million CPU milliseconds. If you've already indexed your queries and moved per-visit writes out of KV, and you still hit a daily limit, that's real traffic, and $5 a month is the cheapest fix. If you haven't, fix the queries first. An unindexed table will keep growing your bill on a paid plan too.