Neon
Neon · databases · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
Compute billed in CU-hours with scale-to-zero after 5 min — $0 for idle, ≈$19.35/mo always-on (0.25 CU) + $0.35/GB storage — and the only official cold-start number in the table: “typically within a few hundred milliseconds”.
#5 of 9 in Databases
ToolScore breakdown — why 78/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸Scale-to-zero after 5 min: idle bills $0 — while an idle VPS bills anyway
- ▸At the hero workload Neon is one of three free tiers (Turso, Neon, Firestore) that genuinely cover the month at $0
- ▸1 CU ≈ 4GB RAM + CPU + SSD; Launch is $0.106/CU-h — the math is published and checkable ($19.35 = 0.25 CU × 730 h × $0.106)
- ▸The only managed provider in the table with an official cold-start number: “typically within a few hundred milliseconds”
- ▸Branching is real — and branch history deltas are billed storage, so it is priced, not hidden
Where it costs you
- ▸Always-on compute is the bill: a 0.25 CU that never sleeps burns 182.5 CU-h/mo ≈ $19.35 before storage
- ▸Cold start “typically within a few hundred milliseconds” (official) — fine for dashboards, risky for p99 APIs
- ▸Branch history deltas are billed storage — every experiment leaves a storage footprint
Hidden traps — the stuff the pricing page doesn't say
After 5 idle minutes the compute parks; the next request waits out a cold start that Neon itself describes as “typically within a few hundred milliseconds”. Fine for dashboards, risky for p99 APIs — the official number is the warning label, use it.
A 0.25 CU that never sleeps burns 182.5 CU-h/mo ≈ $19.35/mo, plus $0.35/GB storage. The $0 you signed up for and the ≈$19.35 you graduate into are the same product in two modes — price the second one on day one.
Branching is the feature; the deltas each branch accumulates are billed as storage. Experiment-heavy workflows grow a storage bill nobody drew on the architecture diagram.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official siteAffiliate link where available — never affects the score.
- official Neon pricing — official, checked 2026-11-12 verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
Postgres on a Hetzner CX23 (2 vCPU / 4GB / 40GB, 20TB traffic) at €5.99/mo ex. VAT — the box that be
The BaaS whose exit is pg_dump — Supabase → any Postgres in 1–5 days, zero schema change — $25/mo Pr
Row-based pricing with the largest free cap in the table — 500M reads + 10M writes/mo covers our her
Neon review
In-depth analysis · last updated 2026-11-12
$0 for idle, published math for everything else
Neon bills compute in CU-hours plus storage, and its defining behavior is scale-to-zero after 5 idle minutes: an idle database bills $0 while an idle VPS bills anyway — the hub’s own “when managed wins” case. At the hero workload (1M reads + 100K writes, ~1GB) Neon is one of the three free tiers (with Turso and Firestore) that genuinely cover the month at $0: 100 CU-h/mo and 1GB of storage on the free plan.
The always-on graduation price
The number to budget is the mode change: always-on compute means a 0.25 CU that never sleeps burns 182.5 CU-h/mo ≈ $19.35, plus $0.35/GB of storage. What makes that number tolerable is that it is published and checkable: 1 CU ≈ 4GB RAM + CPU + SSD, Launch is $0.106/CU-h, and the $19.35 reproduces from the page as 0.25 CU × 730 h × $0.106 — “the math is published and checkable”, the hub’s phrase, and rare in this market.
The honest cold start
Neon is also the only provider here that publishes a cold-start figure: “typically within a few hundred milliseconds” (official). That is the warning label that comes with scale-to-zero — fine for dashboards, risky for p99 APIs. Between the free tier, the branching (branch history deltas are billed storage — priced, not hidden) and the CU abstraction, the product’s shape is unusually legible for this market.
How the pillars land
Price/Value 10: $0 at the hero workload, and an always-on price you can verify with a calculator. Real limits 8.5: 100 CU-h/mo and 1GB free — the workload fits, the absolute caps are small. Latency & cold starts 6.5: the only official number in the table, and it is a few hundred milliseconds you must design around. Data handling sits at the neutral 5 — no data-handling facts in this snapshot. Portability & export 6, neutral band: no export story appears in this snapshot. Feature set 9.5: branching, scale-to-zero, the CU abstraction (1 CU ≈ 4GB RAM + CPU + SSD). Track record 7, neutral band: no track-record facts beyond the end-to-end published math. Ease of use 9.5: zero ops — scale-to-zero after 5 min needs nothing from you. Pricing transparency 10: “the math is published and checkable” — the hub’s line, and earned. Support & docs sits at the neutral 5 — nothing support-related was verifiable from the official pages in this snapshot.
Verdict — SCALE-TO-ZERO PICK, 78/100
Neon is the managed answer for spikey, dashboard-shaped traffic: $0 for idle, published unit math, and a cold start the vendor itself quantifies. If your p99 is the product, the few-hundred-milliseconds restart is the tax; if your traffic is flat and always-on, the $19.35 floor plus storage is the honest number to compare against a VPS.