Comparison · 11 min · 7 citations
Railway vs Render vs Fly Pricing 2026 Compared
Railway vs Render vs Fly pricing 2026: Railway meters usage over a $5 Hobby base, Render sells fixed instances at flat fees, Fly is pure pay-as-you-go.
Railway, Render, and Fly.io now run three different billing models. Railway meters RAM and CPU usage on top of a $5/mo Hobby plan[1]; Render sells fixed-price instances (Starter $7/mo) under flat workspace fees restructured in April 2026[3][4]; Fly.io has no plans at all, just pay-as-you-go[6].
For a solo founder the choice is a risk profile, not a price list: fixed ceiling (Render), hybrid (Railway), or pure meter (Fly). All figures verified against official pricing pages on 2026-07-12.
Comparing these three plan-against-plan misses the point, because one of them has no plans. Fly.io dropped tiered plans for new customers back in October 2024 and bills pure usage. Railway sells a cheap base plan and meters compute on top. Render went the other way in April 2026: flat workspace fees, fixed instance prices, and a bandwidth cut that lands hardest on hobby projects. This comparison puts the three billing models side by side and prices a concrete small app on each.
1. How do Railway, Render, and Fly.io bill in 2026?
| Platform | Plan structure | Compute pricing | Billing model |
|---|---|---|---|
| Railway | Free $0 · Hobby $5/mo (incl. $5 usage) · Pro $20/mo per seat (incl. $20 usage)[1] | Metered per minute: RAM $10/GB/mo, $20/vCPU/mo[1] | Base plan + open-ended meter |
| Render | Hobby $0 · Pro $25/mo flat · Scale $499/mo flat, all with unlimited members[3][4] | Fixed instances, prorated per second: Starter $7 (0.5 CPU/512 MB) up to Pro Ultra $450 (8 CPU/32 GB)[3] | Flat workspace fee + fixed instances |
| Fly.io | No plans; pay-as-you-go for all new customers since October 7, 2024[6] | Per-second Machines, region-priced: shared-cpu-1x 256 MB ~$2.02/mo in Amsterdam[6] | Pure usage meter |
The structural difference matters more than the sticker prices. Render's instance price is a ceiling: a Starter service costs $7 regardless of load[3]. Railway's Hobby plan looks like a ceiling but is a floor, since the $5 includes only $5 of metered usage and everything past that bills at the RAM and CPU rates[1]. Fly has neither floor nor ceiling; you pay for exactly the machine-seconds, storage, and bandwidth you consume[6]. Note also the seat economics: Railway Pro is $20 per seat, while Render's Pro and Scale fees are flat with unlimited members, a real lever the moment you add a collaborator[1][4].
2. Which platform still has a free tier in 2026?
- Railway: a one-time $5 trial credit (30-day expiry, no card required, capped at 1 GB RAM / 2 shared vCPU during trial), then a Free plan that grants a $1 non-rollover credit each month[1][2]. Enough for a tiny service that sleeps most of the time, not for anything always-on.
- Render: free web services at 0.1 CPU / 512 MB that spin down after 15 minutes of idle, 750 free instance-hours per workspace per month, free static sites, and a free Postgres that expires after 30 days[5]. The only ongoing free option here that can serve real traffic, cold starts included.
- Fly.io: no free tier for new accounts. You add a credit card or buy prepaid credits ($25 minimum) before deploying; only pre-sunset accounts keep legacy free allowances[6].
The honest framing for 2026: Render is the last of the three with a usable ongoing free tier, and it cold-starts. Railway's Free plan is a keep-alive for demos. Fly starts charging from the first machine-second. For anything customer-facing, budget the $2 to $7 entry cost and skip the free-tier gymnastics.
3. What does a small always-on app cost on each?
One small always-on web service, no database, modest traffic, platform cost only:
| Path | Spec | Monthly | Annual |
|---|---|---|---|
| Fly.io shared-cpu-1x | 1 shared vCPU / 256 MB | ~$2.02 (Amsterdam rate)[6] | ~$24 |
| Fly.io shared-cpu-2x | 2 shared vCPU / 1 GB | ~$6.64[6] | ~$80 |
| Railway Hobby | 0.5 GB RAM + 0.1 vCPU average draw | ~$7 ($5 base incl. $5 usage + $2 overage)[1] | ~$84 |
| Render Starter | 0.5 CPU / 512 MB, fixed | $7[3] | $84 |
The Railway arithmetic: a service averaging 0.5 GB RAM and 0.1 vCPU meters to about $7/mo (0.5 × $10 RAM + 0.1 × $20 CPU); the Hobby plan's included $5 absorbs the first $5, leaving roughly $2 of overage on top of the $5 base[1]. Fly is the cheapest floor and stays cheap if your app genuinely fits 256 MB, but the meter follows every upsize: extra RAM runs about $5/GB per 30 days[6]. Render costs the same $7 whether the service idles or saturates. Egress diverges too: Railway bills $0.05/GB from the first GB, Render includes 5 GB on Hobby and 25 GB on Pro before $0.15/GB overage, and Fly charges $0.02/GB in North America and Europe[1][3][6]. Fold whichever number wins into your monthly burn rate.
4. Where is the billing risk for a solo founder?
Price tables answer "what does it cost today." This table answers "what can go wrong on the bill":
| Risk dimension | Railway | Render | Fly.io |
|---|---|---|---|
| Bill ceiling | Open-ended meter[1] | Instance price is the ceiling[3] | Open-ended, per second[6] |
| Egress exposure | $0.05/GB, no free allowance beyond plan credit[1] | 5 GB / 25 GB / 1 TB included, then $0.15/GB[3] | $0.02/GB NA/EU, up to $0.12/GB Africa/India[6] |
| Idle behavior | Sleep after ~10 min without traffic; first wake request can fail[7] | Paid instances always on; only free tier spins down[5] | Stopped Machines bill storage only ($0.15/GB per 30 days)[6] |
| Persistent disk | $0.15/GB/mo volumes[1] | $0.25/GB/mo SSD; single-instance attach blocks zero-downtime deploys[3] | $0.15/GB/mo + snapshots $0.08/GB/mo, billable since Jan 1, 2026[6] |
| Team seats | Pro $20 per seat[1] | Flat fee, unlimited members[4] | No seats, pure usage[6] |
| Support / SLA | No email support at any self-serve tier[1] | Contractual SLA Enterprise-only[3] | Support sold separately: $29 / $199 / $2,500+/mo[6] |
| Regions | 4[1] | 5[3] | 18[6] |
Two rows deserve emphasis. Scale-to-zero: Fly is structurally strongest here, since a stopped Machine costs only its root filesystem storage and the proxy restarts it on demand, where Railway's sleep can 502 the first request after idle and Render only sleeps free-tier services[5][6][7]. And support: none of the three bundles guaranteed human support at solo-founder prices, but only Fly lets you buy it outright at $29/mo[6].
5. What changed in 2026?
- Render, April 23, 2026: flat workspace plans replaced per-member pricing (Pro $25 flat vs the old $19/member Professional; Scale $499 flat vs $29/member Organization). Included bandwidth on Hobby fell to 5 GB from the legacy 100 GB, with $0.15/GB overage. Legacy workspaces are force-migrated on August 1, 2026[4].
- Fly.io, January 1, 2026: volume snapshot storage became billable at $0.08/GB/mo past the first 10 GB[6]. Fly's pricing docs also list GPUs as deprecated for new workloads[6].
- Railway: no pricing changes on its changelog in 2026 to date[1]. The current Free plan structure dates from its August 2025 free-plan relaunch.
The Render migration date is the actionable one: if you are on a grandfathered Render workspace, your bill changes shape on August 1, 2026 whether you act or not. Re-check the bandwidth math before then, especially on Hobby.
6. Which should you pick?
- A runaway bill would hurt more than a capped one helps: Render. Fixed instance prices, flat workspace fee, predictable to the dollar[3].
- Lowest floor for a tiny always-on service: Fly.io, ~$2.02/mo, if you watch the meter and your app fits small Machines[6].
- Spiky or scale-to-zero workloads: Fly.io; stopped Machines cost storage only[6].
- Fastest deploy-and-forget for a $5 side project: Railway Hobby, accepting the metered overage past $5 of usage[1].
- Free hobby project that tolerates cold starts: Render's free tier, the only ongoing one left[5].
- Adding teammates soon: Render's flat plans beat Railway's $20/seat as soon as seat two arrives[1][4].
Re-verify each pricing page before committing: two of these three changed billing structure inside the last two years, and the third changed what "free" means. For the database under the app, see Supabase vs Firebase vs Neon; to fold hosting into your full stack cost, use the AI stack cost calculator.
All pricing figures verified against official pricing pages and changelogs as of 2026-07-12.
Frequently asked questions
Does Railway, Render, or Fly.io have a free tier in 2026?
Railway and Render do; Fly.io does not. Railway gives new users a one-time $5 trial credit (30-day expiry, no card), then a Free plan with a $1 non-rollover credit each month, enough for a tiny intermittent service. Render's free web services still exist but spin down after 15 minutes of idle and share 750 free instance-hours per workspace per month, and its free Postgres expires after 30 days plus a 14-day grace period. Fly.io has no free tier for new accounts: you add a credit card or buy prepaid credits with a $25 minimum before anything runs. All three verified against official pricing pages on 2026-07-12.
What changed in Render's April 2026 pricing restructure?
On April 23, 2026 Render replaced per-member pricing with flat workspace fees: Pro is now $25/mo flat (replacing Professional at $19 per member) and Scale is $499/mo flat (replacing Organization at $29 per member), both with unlimited members. The quiet cost: included bandwidth on Hobby dropped to 5 GB, where legacy Hobby included 100 GB, with overage billed at $0.15/GB. Legacy workspaces are force-migrated to the new plans on August 1, 2026, so grandfathered pricing ends then. If your app serves meaningful traffic from a free or Hobby workspace, model the bandwidth line before August.
Is Fly.io cheaper than Railway or Render for a small app?
At the floor, yes: Fly's shared-cpu-1x Machine with 256 MB RAM runs about $2.02/mo in Amsterdam if left on continuously, below Railway's $5 Hobby plan and Render's $7 Starter instance, per the vendors' pricing pages on 2026-07-12. But Fly bills per second with region-specific rates and no plan ceiling, so the bill scales with exactly what you provision: a 1 GB shared-cpu-2x Machine is about $6.64/mo, and extra RAM adds roughly $5/GB per 30 days. Fly also sells reservation blocks at a 40% discount for committed workloads. Cheapest floor, least padded ceiling.
Which billing model is safest for a solo founder?
Fixed-price instances are the safest default because the instance price is the ceiling: on Render, a Starter web service costs $7/mo whether it serves ten requests or ten million, and only bandwidth overage ($0.15/GB past the included allowance) can move the bill. Railway's metered RAM and CPU pricing and Fly's per-second usage billing both have open-ended ceilings, which rewards attention you may not want to spend while shipping. The metered models win when workloads are spiky or scale to zero. If a runaway bill would genuinely hurt, pick the fixed ceiling and revisit once revenue covers the difference.
References
Sources
Primary sources only. No vendor-marketing blogs or aggregated secondary claims.
- 1 Railway — Pricing (Free $0 with $1 monthly credit; Hobby $5/mo incl. $5 usage; Pro $20/seat incl. $20 usage; RAM $10/GB/mo, vCPU $20/mo) — accessed 2026-07-12
- 2 Railway docs — Free trial ($5 one-time credit, 30-day expiry, no card; 1 GB RAM / 2 shared vCPU caps during trial) — accessed 2026-07-12
- 3 Render — Pricing (Hobby $0, Pro $25/mo flat, Scale $499/mo flat; instances Starter $7 to Pro Ultra $450; bandwidth 5 GB/25 GB/1 TB incl., $0.15/GB overage) — accessed 2026-07-12
- 4 Render changelog — Updated plans for Render workspaces (April 23, 2026: flat workspace fees replace per-member fees; legacy workspaces migrate August 1, 2026) — accessed 2026-07-12
- 5 Render docs — Free instances (spin down after 15 min idle; 750 free instance-hours per workspace per month; free Postgres expires after 30 days) — accessed 2026-07-12
- 6 Fly.io docs — Pricing (no plans, pay-as-you-go; shared-cpu-1x 256 MB ~$2.02/mo in Amsterdam; egress from $0.02/GB; volume snapshots billable since Jan 1, 2026) — accessed 2026-07-12
- 7 Railway docs — App sleeping (services sleep after ~10 minutes without outbound traffic; wake on the next inbound request) — accessed 2026-07-12
Tools referenced in this article