Dora NodaDevToolPicks' March 2026 solo-developer guide did something most PaaS comparisons won't: it refused to crown one cheapest platform. Instead it keyed the verdict to your architecture — Railway for low-traffic single-service apps, Render for stable high-traffic ones, Fly.io for workloads that scale to zero. It's the honest answer, and the guide's closing line is worth quoting: Railway wins on price for typical solo workloads, Render wins on predictability, Fly.io wins on latency.
But all three verdicts share one blind spot. Each prices a metered platform against the other metered platforms. None of them prices the same app on a flat-rate box you own, where "cheapest" stops being a property of your architecture and starts being a property of not having a meter at all. So let's recompute each winning architecture — on its home turf, with its own numbers — against a €3.79/month Hetzner CX22 running a self-hosted git-push platform.
The guide's three verdicts, in one table
| Your architecture | Guide's winner | The number it cites |
|---|---|---|
| Low-traffic single-service app (e.g. Laravel SaaS MVP) | Railway | $5–20/mo early-stage |
| Stable, high-traffic app with predictable usage | Render | $7/mo Starter minimum |
| App idle most of the time (scale-to-zero) | Fly.io | ~$1.94/mo machine, plus add-ons |
The guide also names each winner's trap: Railway bills idle containers whether they serve requests or not, Render's free tier sleeps after 15 idle minutes with 30–50 second cold wakes, and Fly.io's sub-$2 machine needs a $2/month dedicated IPv4 plus volumes that bill while stopped. Those traps are where the recompute gets interesting.
The recompute setup: one app, three traffic profiles, four bills
The reference app is the guide's own example shape: a single-service app (Laravel, Rails, or Node) plus Postgres and a small persistent volume. We'll run it under three traffic profiles matching the three verdicts, and price each on all three metered platforms plus the fourth option: a Hetzner CX22 (2 vCPU, 4 GB RAM, 40 GB NVMe, 20 TB of included traffic) at a flat ~€3.79/month — roughly $4.40. One box comfortably holds the app, the database, and half a dozen more side projects. Prices below are the platforms' published 2026 list rates.
Round 1: Railway's home turf — the low-traffic single-service app
Railway's Hobby plan is $5/month including $5 of usage, metered at $10/GB of RAM, $20/vCPU, $0.15/GB-month of volume storage, and $0.05/GB of egress. For a Laravel MVP with 512 MB of app RAM, a modest CPU share, and a Postgres service of similar size, usage lands around $10–15/month against the included $5 credit, so the realistic MVP bill is $10–20/month. That matches the guide's range exactly.
Now the trap the guide flags: idle containers bill. Your staging environment, your idle Postgres, your weekend-dormant side project — all metered, all month. The guide's advice is to enable app sleeping for non-production, which is correct and also an admission: on Railway, the cheapest architecture is the one you remember to turn off.
The fourth option doesn't have an off switch because it doesn't need one. The same app plus Postgres on the CX22 costs €3.79 whether it's serving traffic or sitting idle through a holiday weekend. Staging? Same box, $0 marginal. Second side project? Same box, $0 marginal. Railway's winning architecture — one small service, low traffic — is precisely the shape where a flat-rate box wins by the largest multiple: roughly 3–5x cheaper for the MVP, and the gap widens with every additional idle service.
Round 2: Render's home turf — the stable, high-traffic app
Render's pitch is predictability: a $7/month Starter instance (512 MB RAM, 0.5 CPU) that stays awake, no meter anxiety. The guide is right that for stable high-traffic apps this beats Railway's usage curve — traffic you can forecast favors flat instance pricing over per-GB metering.
But "flat" on Render still has line items. A real backend needs the $7 Starter minimum (the free tier's 15-minute sleep and 30–50 second wake disqualify it for anything user-facing), plus a paid Postgres tier — the free database carries no backup guarantee and expires — plus bandwidth beyond the included quota. A typical app-plus-database setup lands north of $15/month before traffic, and every additional service (worker, staging, second app) is another instance at another $7+.
The CX22 absorbs the same load profile for €3.79 flat. High traffic is the one profile where you'd expect the meter to fight back — 20 TB of included Hetzner traffic versus Render's metered bandwidth — but a solo-dev "high-traffic" app (tens of millions of requests, hundreds of GB of egress) fits inside the box's allowance with room to spare. Render wins its home turf among metered platforms; against the unmetered box it pays a 4x premium for the same always-on posture.
Round 3: Fly.io's home turf — scale to zero
This is the round the fourth option should lose. Fly.io's shared-cpu-1x with 256 MB bills ~$1.94/month running 24/7 — already under the Hetzner box — and auto-stop can take a truly idle machine near zero compute cost. For an app that sleeps 95% of the month, per-second billing is structurally unbeatable by any always-on box.
Except the machine isn't the bill. As the guide notes, a realistic Fly.io app adds a $2/month dedicated IPv4 per app, volumes at $0.15/GB/month billed while the machine is stopped, and egress at $0.02/GB in North America and Europe (up to $0.12/GB in Africa and India). The always-on floor for one small app with an IP and a few GB of volume is ~$4–5/month — already at or above the CX22 — and developers routinely report bills 2–4x their estimates once bandwidth, snapshots, and multi-region copies compound.
So the honest result: Fly.io wins Round 3 only in a narrow band — a single-region app, idle the vast majority of the month, with tiny storage and egress, where you accept cold starts and babysit every add-on. The moment the app has steady traffic, a second region, or a database volume that bills while stopped, the "cheapest for scale-to-zero" verdict converges on the same €3.79 the box charges for everything. Scale-to-zero is the best meter; it's still a meter.
Sensitivity check: when the meter still wins
A recompute that never lets the meter win is propaganda, so here are the boundaries:
- Truly $0 beats everything. Render's free static sites (no sleep penalty) and free-tier backends for portfolio projects cost nothing; a Hetzner box can't beat $0. If your app is a static site or a demo nobody visits, stay on the free tier.
- Near-zero dynamic apps. A personal webhook receiver waking twice a day is genuinely cheaper on Fly.io's per-second billing than any always-on box — roughly $2–3/month (IPv4 plus volume) versus €3.79.
- Multi-region latency. Fly.io's Anycast edge puts your app near users on three continents in minutes. One CX22 in Nuremberg or Ashburn cannot do that at any price; global latency is a product requirement, not a cost line.
- Zero-ops tolerance. The box needs OS patches, Postgres backups, and TLS renewal — minutes a month under a git-push platform, but nonzero. The metered platforms sell freedom from that work, and for some solo devs that's worth $10–15/month.
The real tradeoff was never dollars
Step back and the pattern is clear: DevToolPicks' architecture-keyed verdicts are all correct within the metered world. Railway really is cheapest for low-traffic single-service apps, Render really is cheapest for stable high-traffic ones, and Fly.io really is cheapest near zero. The guide's mistake — the one every three-way PaaS comparison makes — is treating the meter as a law of nature rather than a pricing choice. Once you add the fourth column, "cheapest" stops depending on your traffic shape entirely. It becomes a flat €3.79 that covers all three architectures at once, plus staging, plus the next three side projects.
What the meter actually sells isn't cheapness; it's the absence of ops work. Managed Postgres with backups, zero node maintenance, multi-region in one command — that's a real product worth real money. But price it honestly: you're paying $10–20/month to avoid roughly an hour of platform setup and minutes of monthly maintenance on a box that costs less than a coffee. For many solo developers that's a fine trade. Just don't call it the cheapest option. It never was.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.
Sources
- Railway vs Render vs Fly.io for Solo Developers in 2026 — DevToolPicks, March 27, 2026: architecture-keyed verdicts, $5–20/mo Laravel MVP, idle-container billing warning
- Railway pricing reference via 2026 community rate cards — Hobby $5/mo incl. $5 usage; RAM $10/GB/mo, CPU $20/vCPU/mo, volumes $0.15/GB/mo, egress $0.05/GB
- Render pricing via 2026 deployment guides — free spin-down after 15 min idle with 30–50s wake; Starter $7/mo always-on (512 MB/0.5 CPU)
- Fly.io pricing via 2026 rate cards — shared-cpu-1x 256 MB ~$1.94/mo, dedicated IPv4 $2/mo, volumes $0.15/GB/mo billed while stopped, egress $0.02–0.12/GB by region
- Hetzner vs Vultr post–June 2026 comparison and Hetzner–Vultr benchmarks — CX22 €3.79/mo (2 vCPU/4 GB/40 GB/20 TB); June 15, 2026 repricing concentrated on CPX/CCX lines
- Railway vs Render vs Fly.io: Benchmarks & Pricing (2026) — Techsy cross-check on Fly.io egress tiers and Render's April 2026 plan restructure
