Neon vs PlanetScale 2026: Postgres Pricing Compared at Four Sizes

Choose Neon for a database that is idle most of the day or that needs branches: the Free plan is real, compute scales to zero after five minutes, and Launch bills $0.106 per compute-unit hour. Choose PlanetScale for a production Postgres that runs all month and must survive a node failure: a three-node PS-10 cluster with 1 GiB of RAM costs $30 a month on ARM, and there is no free plan, only a single-node entry instance.

Prices below were read on Neon's pricing page and plan docs and on PlanetScale's pricing page and billing docs on 17 October 2026. The two products sell Postgres in opposite shapes. Neon meters hours and gigabytes on an elastic compute; PlanetScale sells fixed instances, prorated to the millisecond, with replicas and NVMe as the upsell. That difference, not the feature list, decides which is cheaper for you.

Toolradar data: the managed databases ranking we run evaluated 24 tools, updated October 2026. Among the 53 serverless tools in the directory, PlanetScale is listed as an alternative to 17 others and Neon to 12, and visitors to the PlanetScale page compare it with Neon more than with any other product, as of October 2026.

At a glance

A managed Postgres, for a developer, is a Postgres server someone else patches, backs up and restarts, with a connection string you paste into an app. Both vendors deliver that. Neon adds a storage and compute split that lets the database sleep and fork; PlanetScale adds the operational posture of a database company that spent years running Vitess for MySQL and now applies it to Postgres.

The single-product pages are the Neon review and the PlanetScale review. For the wider field, best Postgres hosting ranks both against Supabase, Crunchy and the clouds, best backend-as-a-service covers the bundled option, and Neon vs Supabase prices Neon against the other elastic option.

Costed scenarios, one database with 10 GB of data

Each total multiplies the vendors' published rates for one database holding 10 GB of data, which PlanetScale includes with every cluster and Neon bills at $0.35 per GB-month. A month is 730 hours. None of these is a quote.

Two things stand out. At matched memory and always-on, the two land within a few dollars of each other, and the PlanetScale figure buys a primary and two replicas while the Neon figure buys one compute. That makes PlanetScale the better deal for a production database that cannot pause, because the same money includes failover. The moment the database idles, Neon pulls away. A 4 GB compute awake a third of the day costs about a third of the always-on bill, and PlanetScale has no equivalent lever.

How Neon bills: compute units, storage and the sleep timer

Neon's unit is the compute unit, which its docs describe as roughly 4 GB of RAM with proportional CPU, so 0.25 CU is 1 GB and 2 CU is 8 GB. Launch charges $0.106 per CU-hour and Scale $0.222 per CU-hour, with no monthly minimum on either; the Scale rate buys 1,000 projects, 25 branches, 30 days of restore history and standard support rather than more hardware. Storage is $0.35 per GB-month on both, with 500 GB of network transfer included per project and $0.10 per GB after.

Autoscaling runs between 0.25 and 16 CU, with at most an 8 CU span between the minimum and maximum you set, so a database can sit at 0.25 CU overnight and climb to 2 CU under load. Scale to zero suspends the compute after 5 minutes of inactivity and wakes it in a few hundred milliseconds on the next connection; paid plans can switch it off, Free cannot. Instant restore costs $0.20 per GB-month of history on paid plans, up to 7 days on Launch and 30 on Scale.

The Free plan is a working database rather than a demo: 100 CU-hours a month per project, 1 GB of storage per project up to 20 GB across the account, 100 projects, 10 branches each and computes up to 2 CU. When a project uses its hours, compute stays suspended until the next cycle. A side project at 0.25 CU that is awake for 400 hours in the month fits inside the allowance.

How PlanetScale bills: instances, nodes and NVMe

PlanetScale Postgres is a menu of fixed instances. A single node PS-5 (1/16 vCPU, 512 MiB) is $5 a month and PS-10 (1/8 vCPU, 1 GiB) is $10 to $13 depending on architecture. The production shape is three nodes, a primary with two replicas, where PS-10 costs $30 on ARM or $39 on x86-64, PS-20 with 2 GiB is $50 or $59, PS-40 with 4 GiB is $83 or $99 and PS-80 with 8 GiB is $148 or $179. The ladder continues to PS-2560 with 256 GiB at $4,529 to $5,599.

Billing is prorated to the millisecond, so a cluster created on the 20th pays a third of a month. Each cluster includes its first 10 GB of network-attached storage, backups equal to twice the disk size at no charge, 100 GB of public egress on production branches and free inbound traffic. Extra storage is billed per region, additional replicas cost about a third of the HA rate, cross-region replica traffic runs $0.02 to $0.16 per GB and private connections over AWS PrivateLink or Google Private Service Connect cost $0.01 per GB.

Metal is the second ladder: three-node clusters with local NVMe where storage is bundled into the instance price, from M-10 at $50 a month to configurations above $47,000. It is the option for a write-heavy database where network storage latency shows up in the p99, and it has no equivalent on Neon's price list.

Where each one wins

Neon wins on idle time. Every hour the database sleeps is an hour you do not pay for, and the sleep is automatic. Staging copies, preview databases per pull request, internal tools used in office hours and side projects all cost a fraction of a fixed instance.

Neon wins on branching. Ten branches are included on Launch and each extra one is $1.50 a month, which is cheap enough to give every feature branch its own database. PlanetScale's Postgres pricing page does not price branches as a line item, so the workflow exists there without the same per-branch cost model.

PlanetScale wins on always-on production. At 1, 4 and 8 GB the three-node cluster costs about what one Neon compute costs, and it includes two replicas and automatic failover. A database that serves traffic 24 hours a day gets more hardware per dollar.

PlanetScale wins on ceiling and I/O. The instance ladder runs to 256 GiB on network storage and further on Metal with local NVMe. Neon's autoscaling stops at 16 CU, or 64 GB of RAM, and larger fixed computes lose scale to zero.

Neither has a free lunch at the bottom. Neon's Free plan suspends when the 100 CU-hours run out and PlanetScale's cheapest node still costs money every month, with no free plan at all. A prototype that must stay reachable every minute of the month is cheaper on the PS-5 than on a Neon Launch compute that never sleeps.

The traps on each invoice

Neon's trap is a compute that never idles. Disable scale to zero, or run a connection pool that keeps the database awake, and a 1 GB compute costs about $19 a month in compute alone on Launch, more than a PlanetScale single node with the same memory. The second trap is branch sprawl at $1.50 each above the included count, which is small per branch and large per team by quarter end.

PlanetScale's trap is architecture and replicas. The x86-64 price is 20 to 30 percent above ARM at every size, and a replica added beyond the two included costs about a third of the HA rate again. Storage above the included 10 GB and egress above 100 GB are billed at regional rates the pricing page does not print, so a data-heavy workload should price those with PlanetScale's rate tables before committing.

Which should you choose?

Pick Neon for anything that idles, forks or is still being built: side projects, preview environments, internal tools, a startup's first production database while traffic is low. Start on Free, move to Launch when the hours run out, and only move to Scale for the longer restore window and the higher project count.

Pick PlanetScale for a production Postgres with steady traffic that needs failover now: the three-node PS-10 on ARM is the cheapest highly available Postgres on either price list, and the ladder above it is predictable. Pick Metal when NVMe latency matters more than the monthly bill.

Pick Supabase instead of either if you want auth, storage and edge functions on the same invoice; Neon vs Supabase and the Supabase pricing breakdown cover that choice, and best free database hosting lists the free tiers if cost is the only question. For the client you connect with, see best database clients.

How we compared

We read Neon's pricing page, plans doc, compute doc and scale-to-zero doc, plus PlanetScale's pricing page and Postgres billing doc, on 17 October 2026. The scenario totals multiply those published rates: CU-hours times the Launch rate plus 10 GB of storage on Neon, and the listed instance price on PlanetScale with its included 10 GB. Cross-shopping figures come from Toolradar, the software directory Dupple runs. We run no benchmark lab, so this page makes no latency or throughput claim. Louis Corneloup, founder of Toolradar and Dupple, edited the verdict; editorial rules are on the Devshot about page.

FAQ

Which is cheaper in 2026, Neon or PlanetScale?

It depends on whether the database idles. Always on at 4 GB of memory, Neon Launch costs about $80.88 a month with 10 GB of data and PlanetScale's three-node PS-40 costs $83 on ARM, so they are level and PlanetScale includes two replicas. Awake eight hours a day, Neon drops to about $28.94 while PlanetScale stays at $83. For a prototype, Neon's Free plan beats PlanetScale's cheapest single node.

Does PlanetScale have a free plan?

No. PlanetScale's pricing page lists no free tier for Postgres; the cheapest option is a PS-5 single node at $5 a month with 1/16 vCPU and 512 MiB of RAM, and the cheapest highly available cluster is the PS-5 with a primary and two replicas at $15. Billing is prorated to the millisecond, so a short experiment costs a fraction of the monthly price.

How does Neon's scale to zero work?

Neon suspends a compute after 5 minutes without activity and restarts it in a few hundred milliseconds when a connection arrives. You pay only for the hours the compute runs, at the Launch per-CU-hour rate. Paid plans can disable the behaviour for an always-on database; the Free plan cannot, and computes above 16 CU stay always active.

Is PlanetScale Postgres highly available by default?

The production shape is a three-node cluster, one primary and two replicas, which is what the PS-10 and PS-40 ARM prices in the scenario table buy. Single-node instances are sold at roughly a third of the HA price for development use. Backups equal to twice the disk size are included at no charge, and extra replicas cost about a third of the HA rate each.

What does a Neon compute unit cost?

Launch charges $0.106 per CU-hour and Scale $0.222, where a compute unit is roughly 4 GB of RAM with proportional CPU. A 1 CU compute running all month is 730 CU-hours, about $77 on Launch before storage at $0.35 per GB-month. Autoscaling lets a database move between 0.25 and 16 CU, so the hourly rate applies to the size actually in use.

Can I run MySQL on either?

PlanetScale yes: its Vitess-based MySQL clusters start at PS-10 for $39 a month on three nodes and run to PS-2800, with Metal MySQL and the Neki sharding platform above that. Neon is Postgres only. A team choosing between the two for MySQL has no decision to make; the comparison on this page is Postgres against Postgres.

Cite this: Devshot, "Neon vs PlanetScale 2026: Postgres Pricing Compared at Four Sizes", October 2026.

Devshot is the daily brief on tools and changes developers act on. Join at devshot.co.

Some offers on this page may be paid placements or contain affiliate links.