PaaS pricing is the sum of three separate things: a plan or seat fee, a compute charge for each running process, and metered usage such as bandwidth and database storage. Vercel, Render, Heroku, Railway and Fly.io each combine those three differently, which is why the same app can cost $54 or $109 a month depending on where it runs.
You have a small production app: a web process, a background worker, and a Postgres database with a few gigabytes in it. Every pricing page answers a different question: Heroku quotes a monthly maximum per dyno, Render an instance price plus a workspace tier, Vercel a per-seat fee and a table of usage lines, Railway cents per vCPU-second. None of those numbers is wrong. They are just not comparable until you know what each includes.
How PaaS Pricing Actually Works
Every PaaS bill has the same three layers: a platform fee, compute, and metered usage. Plan tables show only the middle one, and the platforms disagree on what “running” even means for it: allocated, awake, or actively using CPU.
Four billing shapes
Heroku bills wall-clock dyno time with a monthly cap. Usage comes “from wall-clock time, not CPU time,” any dyno scaled above zero accrues it, prorated to the second, and the published monthly price is what you pay if the dyno runs all cycle.[1] An idle Standard-1X costs the same $25 as a busy one.
Render bills a workspace fee plus per-second instances. You pick a workspace tier (Hobby at $0, Pro at $25, Scale at $499) that sets team features and included bandwidth, then each service is a fixed instance size prorated to the second.[2]
Vercel bills per seat plus usage credit plus fluid compute. The Pro platform fee is $20 a month, carrying one deploying seat and $20 of team usage credit; everything past it is on-demand: data transfer, edge requests, invocations, and compute measured as Active CPU hours and provisioned memory GB-hours at regional rates.[3]
Railway and Fly.io bill per-second resource usage. Railway meters memory and vCPU actually consumed ($10 per GB and $20 per vCPU, monthly equivalents) on top of a $5 or $20 plan that includes the same amount of usage, and stopped services cost nothing.[4] Fly.io prices a named CPU and RAM preset per second while a Machine runs, and only its root filesystem while it is stopped, with no plan fee.
How Much Does Vercel Cost?
Vercel costs $0 on Hobby for personal, non-commercial projects, and $20 a month on Pro for the platform fee, which includes one deploying seat and $20 of usage credit.[3] Extra deploying seats are $20 each; viewer seats are read-only and cost nothing.[5]
Pro includes 1 TB of fast data transfer, then from $0.15 per GB, and 10 million edge requests, then from $2 per million. Invocations are $0.60 per million, and compute is regional: Active CPU is $0.128 an hour in Washington (iad1) against $0.177 in Paris or San Francisco, with provisioned memory at $0.0106 per GB-hour in iad1.[6]
The unit of compute is a function instance, paused once the last in-flight request finishes and charged nothing until the next invocation: right for a Next.js frontend, wrong for a queue consumer. And Vercel sells no database of its own; Postgres, Redis and vector stores come from Marketplace providers such as Neon, Upstash and Supabase, at their prices.[7]
When does Vercel get expensive?
Vercel gets expensive when metered usage outgrows the $20 credit, and the lines that outrun it first are the ones traffic drives: transfer past the included terabyte, edge requests past 10 million, and invocations. Compute holds up better under growth, because Active CPU pauses while a function waits on I/O. Seats and add-ons sit outside the credit, so headcount and a $100-a-month Static IP are additive on top.
The control is Spend Management. Set an amount per billing cycle and Vercel can notify you at 50, 75 and 100 percent, POST to a webhook, or pause production deployments across every project.[8] The check runs every few minutes rather than continuously, so its docs advise setting the amount below the maximum you would tolerate.
How Much Does Render Cost?
Render costs a workspace fee plus the compute you run. Hobby is $0 a month with 5 GB of bandwidth and up to 25 services; Pro is $25 with unlimited seats and services and 25 GB; Scale is $499 with 1 TB. Bandwidth over the allowance is $0.15 per GB on every tier,[2] and the workspace tier is the first thing a startup picking a host has to choose.
Web services and workers are fixed instances. The smallest paid instance is $7 a month for 512 MB and less than a full CPU; 1 CPU with 2 GB is $25 and 2 CPU with 4 GB is $85. Persistent disks are $0.25 per GB per month.
Render Postgres starts at $6 a month for 256 MB of RAM, then $19 for 1 GB, $40 for 1 CPU and 2 GB, and $75 for 2 CPU and 4 GB, each with 1 GB of SSD included and expandable storage at $0.30 per GB. The best Postgres hosting comparison covers what those prices should include.
The catch people search for is the $0 instance: Render spins one down after 15 minutes without inbound traffic and takes about a minute to bring it back, grants each workspace 750 Free instance hours a calendar month, and expires a Free Postgres 30 days after creation.[9]
How Much Does Heroku Cost?
Heroku costs $5 a month at minimum, for the Eco plan, and there is no plan fee above that: you pay per dyno, per database, and per add-on. Eco buys 1,000 dyno hours shared across every Eco dyno in the account,[1] and those dynos sleep after 30 minutes of inactivity. Basic is $7 for an always-on 512 MB dyno. Standard-1X is $25 for 512 MB, Standard-2X $50 for 1 GB, Performance-M $250 for 2.5 GB and Performance-L $500 for 14 GB.[10]
Every figure is a monthly maximum: a dyno scaled to zero for half the month costs half, and one left running costs the full amount whether it serves one request or a million. No per-request component and no workspace tier is why Heroku bills are predictable.
Heroku Postgres runs $5 for Essential-0 with 1 GB of storage, $9 for Essential-1 with 10 GB and $20 for Essential-2 with 32 GB; production Standard plans start at $50.
On February 6, 2026, Heroku announced a move to a sustaining-engineering model: credit-card customers, existing and new, see no change to pricing or billing, but Enterprise Account contracts are no longer offered to new customers.[11] Where teams go from there is the Heroku alternatives question.
The Same App Priced on All Five
A small production app (one always-on web process, one worker, one Postgres with about 10 GB of disk) lands between about $54 and $82 a month on the four platforms that can run all of it; Vercel runs only the web tier. The sizes are each platform’s smallest production-grade ones, and the RAM is shown because that is where comparisons diverge.
| Platform | Web | Worker | Postgres | Plan fee | Monthly total |
|---|---|---|---|---|---|
| Heroku | Standard-1X, 512 MB, $25 | Standard-1X, 512 MB, $25 | Essential-1, 10 GB, $9 | none | $59 |
| Render | 1 CPU, 2 GB, $25 | 0.5 CPU, 512 MB, $7 | 1 GB RAM, $19, plus 9 GB at $0.30 ($2.70) | $0 Hobby or $25 Pro | $53.70 to $78.70 |
| Vercel | Pro, $20 platform fee, usage-billed | none offered | Marketplace provider | $20 with one seat | $20 plus usage |
| Railway | 1 vCPU, 2 GB, up to $40 | 0.5 vCPU, 1 GB, up to $20 | 1 GB, 0.5 vCPU, up to $20, plus 10 GB volume ($1.50) | $5 Hobby (includes $5) | up to $81.50, less when idle |
| Fly.io (iad) | shared-cpu-1x, 2 GB, $10.70 | shared-cpu-1x, 1 GB, $5.70 | Managed Postgres Basic, 1 GB, $38, plus 10 GB at $0.28 ($2.80) | none | $57.20 |
Totals are arithmetic on the published rates, not figures any vendor quotes. Rates checked 2026-09-10.[1][2][3][4][10][12][13] Railway figures assume the allocation is fully used all month; it bills actual consumption per second. Heroku figures are monthly maximums.
Heroku’s $59 buys the least memory, 1 GB across both processes; moving both to Standard-2X takes it to $109. Render’s compute is the cheapest per gigabyte, but the $25 Pro workspace a team of two effectively needs is a fixed cost the others lack. Fly.io inverts the shape: $16.40 of compute for 3 GB, and the most expensive database line, because every Managed Postgres plan includes high availability, backups and connection pooling.
Is a Fly shared-cpu-1x the same as a Render Standard or Heroku Std-1X?
No, and Fly.io’s own CPU documentation explains why: a shared vCPU is guaranteed 5 ms of every 80 ms period, 6.25 percent of a physical core, and runs at up to 100 percent only while it has burst balance. Fine for a web process that mostly waits on the database; wrong for anything CPU-bound. The honest like-for-like for a busy service is performance-1x, a full core, at $31 a month with 2 GB in Ashburn. Swap it in and Fly.io’s total is $77.50, Render-on-Pro territory rather than half of it.
PaaS Pricing on Fly.io
Fly.io has one billing shape: every Machine is a named CPU and RAM preset billed per second while it runs, from $1.94 a month for a shared-cpu-1x with 256 MB in Ashburn, with additional RAM at about $5 per GB per 30 days. A stopped Machine costs $0.15 per GB of root filesystem per month, so a worker that runs an hour a day is billed for one. There is no workspace tier and no per-seat charge. The pricing page lists every rate by region, and the calculator prices a web, worker and Postgres from those rates before you deploy.
Managed Postgres is priced by plan and storage: Basic is $38 a month for 1 GB of RAM, Starter $72 for 2 GB, and storage $0.28 per provisioned GB. Volumes are $0.15 per GB, and egress from North America and Europe is $0.02 per GB; inbound and same-region private traffic are not billed.
The sizing caveat is real, and the CPU performance docs put numbers on it: use shared-cpu for I/O-bound processes and performance-cpu where you would have bought a full core elsewhere. Either way the meter runs by the second, and a Machine that stops on idle stops costing compute. That same model prices sandboxes for agents, which AI sandbox pricing compares on its own terms.
Frequently Asked Questions
Is Render better than Heroku on price?
Yes, usually, at small sizes: Render’s 1 CPU, 2 GB instance is $25 against Heroku’s $25 Standard-1X with 512 MB. Heroku wins on Postgres entry pricing ($5 and $9) and has no workspace fee. Fly.io prices that shape per second: a shared-cpu-1x with 2 GB is $10.70 in Ashburn.
Is Render or Vercel cheaper?
For a frontend with light API routes, Vercel: $20 covers the platform fee, one deploying seat and $20 of usage credit. For an always-on server, a worker or a database, Render, because Vercel’s compute is request-scoped and its databases come from Marketplace providers. Fly.io sits with Render, on per-second Machines with no per-seat charge.
Which is better, Vercel or Heroku?
They solve different problems. Heroku runs full applications on always-on dynos with a fixed monthly maximum; Vercel runs frontends and functions on usage-billed fluid compute. Heroku has been in a sustaining-engineering model since February 2026. Fly.io runs both shapes on the same per-second Machines.
Is Heroku still free?
No. The cheapest dyno today is the Eco plan, $5 a month for 1,000 dyno hours shared across the account, and Eco dynos sleep after 30 minutes of inactivity. On Fly.io a shared-cpu-1x Machine is $1.94 a month in Ashburn, billed per second.
Is Render free or paid?
Both. The Hobby workspace is $0 and Render’s Free instances cost nothing, but they spin down after 15 minutes without inbound traffic, a workspace gets 750 Free instance hours a calendar month, and a Free Postgres expires 30 days after creation. Fly.io bills every Machine per second on published rates.
Is a Fly.io shared-cpu-1x the same as a Render Standard or Heroku Standard-1X?
No. A Fly.io shared vCPU is guaranteed 6.25 percent of a physical core (5 ms in every 80 ms) plus a burst balance for peaks, so a $10.70 shared-cpu-1x is not equivalent to Render’s $25 full-CPU instance. The like-for-like Fly.io size for a busy service is performance-1x, $31 a month with 2 GB.
Sources
- ^ Heroku Dev Center, “Usage and Billing”. devcenter.heroku.com/articles/usage-and-billing. Checked 2026-09-10.
- ^ Render, “Pricing”. render.com/pricing. Checked 2026-09-10.
- ^ Vercel, “Pricing”. vercel.com/pricing. Checked 2026-09-10.
- ^ Railway, “Pricing”. railway.com/pricing. Checked 2026-09-10.
- ^ Vercel, “Vercel Pro Plan”. vercel.com/docs/plans/pro-plan. Checked 2026-09-10.
- ^ Vercel, “Fluid compute pricing”. vercel.com/docs/functions/usage-and-pricing. Checked 2026-09-10.
- ^ Vercel, “Vercel Storage overview”. vercel.com/docs/storage. Checked 2026-09-10.
- ^ Vercel, “Spend Management”. vercel.com/docs/spend-management. Checked 2026-09-10.
- ^ Render, “Free Instances”. render.com/docs/free. Checked 2026-09-10.
- ^ Heroku, “Pricing”. heroku.com/pricing. Checked 2026-09-10.
- ^ Heroku, “An Update on Heroku”. heroku.com/blog/an-update-on-heroku. Checked 2026-09-10.
- ^ Fly.io, “Pricing”. fly.io/docs/about/pricing. Checked 2026-09-10.
- ^ Fly.io, “Managed Postgres”. fly.io/docs/mpg. Checked 2026-09-10.