MongoDB Atlas free tier limits: 512 MB and 100 operations per second

MongoDB Atlas has given away a cluster for years, and the terms have moved: shared hardware, an operations-per-second ceiling, a sort-memory cap, a connection limit buried in the documentation rather than the pricing page. This page records what the free tier actually permits on 2026-10-25 and what the first invoice looks like when a project outgrows it.

The free tier is the M0 cluster: 512 MB of storage, shared RAM and shared vCPU, at $0 per hour and listed as free forever. The first paid step is Flex at $0.011 per hour, which MongoDB prices at $8.00 a month in the lowest throughput band and caps at $30 a month.

What the free tier actually limits

The two ceilings printed in the pricing FAQ are 100 operations per second and 32 MB of sort memory. The documentation adds a third: a free cluster can hold at most 500 connections, and once the queue behind the operations-per-second limit grows, operations wait more than a second before they run.

What the numbers actually mean

Operations per second is the number to watch, because it is not requests per second. A query that scans a collection counts as many operations, so a workload with the right index stays under 100 far longer than the same traffic doing ad-hoc aggregation.

Flex is metered the same way but sold in bands. The pricing page lists five: 0-100 operations per second at $0.0110 per hour, then $0.0205, $0.0288, $0.0356 and $0.0411 as the ceiling rises by 100 each time. Run those rates over 730 hours and the bands map to $8.00, $15.00, $21.00, $26.00 and $30.00, which is where the "up to $30/month" on the Flex card comes from.

Storage is the other axis, and the first paid step is bigger than it looks. Flex includes 5 GB. The M2 free-cluster tier includes 2 GB for $9 a month; M5 includes 5 GB for $25. Against a 512 MB free cluster, M2 is a fourfold increase in space and M5 a tenfold one.

How we measured it

Every figure here is copied from the Atlas pricing page and the free-cluster limitations page, both opened on 2026-10-25. Nothing was provisioned and no benchmark was run, so these are list prices before any commitment discount. Monthly Flex figures use the page's own assumption of continuous usage for 30 days, the same arithmetic it prints beside each band, and are labelled calculated where they are derived rather than quoted.

What changes in practice

The gap between free and paid is where projects get hurt. M0 is not a small version of a paid cluster; it is a cluster with a throughput ceiling, a sort ceiling and a connection ceiling that never appear on a capacity chart until you hit one. The first symptom is usually latency during a deploy or a cron job, not an error.

The connection limit bites serverless deployments hardest. 500 connections sounds generous until each function invocation opens its own, and the documentation's warning about operations waiting behind the per-second queue describes a real stall rather than a slow query plan.

Flex is billed hourly, so the band you occupy during a spike is the band you pay for that hour. That makes it a reasonable home for a staging database and a poor home for a batch job you can only run at 4 a.m., where the same work on a dedicated tier bills at a known rate.

M5, at 5 GB for $25 a month, undercuts Flex's top two bands of $26 and $30 while including the same 5 GB, which makes a fixed shared cluster the cheaper home for a database whose throughput has stopped moving even as its storage grows.

Bottom line

Under 512 MB of data and 100 operations per second, Atlas costs nothing and there is no case for paying. The moment indexes stop fitting that throughput budget, budget $9 a month for M2 and treat $30 as the Flex ceiling that decides whether you stay on shared hardware or move to a dedicated M10 at $56.94 a month.