Dora NodaOn February 6, 2026, Heroku told customers it was entering sustaining-engineering mode: stability, security, and reliability patches only, no new features, and no new Enterprise contracts. Existing apps keep running, but Salesforce has stopped selling the platform's future. Ten days later, Back4App published a ranking that, intentionally or not, answers the only question that announcement left open — if not Heroku, what?
Back4App's February 2026 guide ranks 10 Heroku alternatives by pricing, free tier, and migration fit: Back4App, Sevalla, Render, Fly.io, DigitalOcean App Platform, Railway, Vercel, Upsun, Coolify, and AWS Elastic Beanstalk. It is the rare third-party comparison that lists a self-hosted tool (Coolify) alongside managed platforms and names Heroku's sustaining mode as the forcing function, rather than treating it as a footnote.
This post re-audits that ranking line by line against the number every 2026 PaaS comparison eventually circles back to: a flat-rate Hetzner CX22 — 2 vCPU, 4 GB RAM, 40 GB SSD, 20 TB included traffic — now €4.49/month after Hetzner's June 2026 repricing. The next table states the answer before any narrative smoothing, and the rest of this post shows where each number came from.
The answer up front: 10 alternatives, one flat box
The workload is deliberately ordinary — what a small team actually migrates off Heroku:
- App — Node.js or Python web service
- Worker — one background worker (same image, different process)
- Database — one Postgres, small production dataset
Pricing is from public pages verified through mid-2026. Hetzner pricing is the post-June-15 table (EU, ex-VAT, hourly capped to monthly). Hosted DB pricing uses the cheapest managed Postgres each platform actually offers; self-hosted counts Postgres as a co-located workload on the same box.
| # | Platform (Back4App order) | Free tier | Entry paid | Bandwidth / egress model | What it actually competes on | Where the meter bites vs a €4.49 CX22 |
|---|---|---|---|---|---|---|
| 1 | Back4App (BaaS) | Free tier: 25K requests/mo, 1 GB storage | Shared from ~$5/mo; dedicated from ~$15/mo; enterprise custom | Included in plan tiers; overage per resource, not per-GB egress | Parse-based BaaS — backend + DB + APIs without writing the backend | Cheaper than a CX22 for a Parse-shaped app; more expensive the moment you outgrow the BaaS data model |
| 2 | Sevalla | Free tier available | From ~$7-15/mo depending on resource class | Included per plan; scales with resource tier | Managed PaaS with a Heroku-like deploy model at a lower entry price | Undercuts Heroku's old floor; still metered — two services exceed one CX22 on fixed cost |
| 3 | Render | Yes — 750 hours/mo (services sleep; Postgres free tier is 30-day trial, not permanent) | Starter $7/mo per service | 100 GB included, then $0.30/GB overage (egress-heavy apps hit this first) | Closest Heroku UX match with a real free tier | Free tier beats Hetzner at $0; one always-on service + DB already costs more than a CX22 before egress |
| 4 | Fly.io | No — 2-hour trial only, then pay from first dollar (Hobby/Launch/Scale gone since Oct 2024) | ~$1.94-5.92/mo per Machine (256 MB to 1 GB) + volumes + egress | $0.02/GB NA/EU, $0.12/GB Africa/India — 6x regional multiplier | Multi-region edge with per-second billing; cheapest always-on single service in NA/EU | Cheapest at low egress and one region; 3-service prod stack with 200 GB egress already ~$18-28 vs one CX22 |
| 5 | DigitalOcean App Platform | Free tier for static sites; paid for containers | Basic $5/mo per service, Pro from higher tiers | 100 GB included on paid tiers; bandwidth pool scales with tier | DO-native deploy with managed infra primitives | Small stack competitive; per-service multiplication pushes past CX22 quickly |
| 6 | Railway | Not permanent — $1/mo minimum after 30-day trial (trial credits, not a free tier) | Hobby $5/mo, Pro $20/mo (both are credit pools, not caps) | Metered egress on top of compute credit pool | Fastest DX, built-in managed Postgres, good CLI | Hobby floor is actually $5-20/mo for the credit pool before egress; 3-service stack is 3-5x a CX22 |
| 7 | Vercel | Hobby free (generous for frontend) | Pro $20/user/mo + usage | 1 TB included on Pro, then $0.15/GB ($40/100 GB) after 2026 bandwidth hike | Frontend PaaS — best for Next.js/SSR, not general containers | Bandwidth overage is the whole story: media-heavy apps overtake a CX22 on egress alone |
| 8 | Upsun (Platform.sh successor) | No free tier | Per-project orchestration fee + per-hour CPU/memory + per-GB storage/backups across 4 billed dimensions, on AWS/GCP/Azure/OVH/IBM | Dimension-based, not per-GB egress — storage and env count drive cost | Choose-your-cloud per app; true multi-cloud flexibility | Four-dimensional billing makes even a small multi-service app total more than one CX22; multi-cloud is insurance most teams don't need |
| 9 | Coolify | Free (self-hosted) — you bring the server | $5/mo Coolify Cloud if you want it managed, otherwise $0 on your box | Bandwidth is the server's — Hetzner's 20 TB if you host there | Self-hosted Heroku replacement with a web UI | The only entry that collapses to the CX22 price — because it is the CX22 with a different dashboard |
| 10 | AWS Elastic Beanstalk | Part of AWS Free Tier (12 months) | Pay for underlying EC2/RDS/etc. — no PaaS fee, but no PaaS abstraction either | EC2 egress rates apply | Least opinionated — you own the infra shape | Cheapest only if you size raw EC2 small; operational cost is the hidden line |
Three things this table makes visible that the ranking's own green checkmarks don't:
First, "free tier" means four different things in the same list — permanent free (Render Hobby), 30-day trial (Render Postgres, Railway), 2-hour trial (Fly.io), and self-hosted-free (Coolify) — but the comparison renders them as equally green. A team that migrates assuming "free" is a steady state will hit a wall on day 31, not day 730.
Second, the egress row is the actual decision, not the compute row. Every platform's compute entry price is within a few dollars of every other. Egress — $0.02 vs $0.12 vs $0.30 vs 20 TB included — is where a $5/mo service becomes a $50/mo invoice.
Third, only one entry has no meter at all. A CX22 costs €4.49 whether you ship 10 GB or 10 TB that month (up to the 20 TB included). Every other row adds at least one meter the tenant can't cap except by shipping fewer bytes.
The rest of this post walks each group in that table — the cheap-looking three, the differently-priced three, and the platform whose non-price is its price.
Why a third-party ranking is a better anchor than any vendor's pricing page
Single-vendor pricing pages are optimized to answer "how cheap can your first month be." Back4App's ranking is accidentally optimized for the harder question — "what will the bill look like when you're actually running the app you migrated?"
Three choices in its framing reveal costs the vendors bury:
Coolify on the list changes the question. Most Heroku-alternatives roundups enumerate only other hosted platforms, which frames "where to run the app next" as "which other vendor." Including Coolify — an open-source, self-hosted PaaS you run on your own VPS — forces the build-vs-buy comparison onto the same page as the buy-vs-buy one. That is the only placement where the CX22 number can appear as a like-for-like entry rather than a separate blog post's rebuttal. It also surfaces the tradeoff the ranking's copy is honest about: Coolify is free to run, but the team that picks it owns backups, upgrades, and on-call for the box that runs it.
Heroku as sustaining mode, not failed platform, reframes the timing. The ranking doesn't argue Heroku is down or broken; it names the category — sustaining engineering, no new Enterprise contracts since February 6 — and lets the reader price the implication themselves. That's more useful than a "Heroku is dying" hot take, because the actual cost of staying on a sustaining-mode platform isn't downtime, it's lost leverage: no new features to plan against, no enterprise negotiation, and migration debt that compounds the longer you defer the decision. The right question after reading it is not "is Heroku dead" but "what does staying cost me next year versus moving now."
Migration notes as a pricing signal. Back4App annotates each alternative with what migrating to it entails — buildpack compatibility, database import paths, add-on replacement. Those notes double as a cost estimate: a platform that accepts your Heroku buildpacks and Postgres dump is cheaper to migrate to than one that wants you to rewrite the app, even if the entry price looks similar. No vendor's own pricing page will tell you its migration is the expensive one.
None of this makes the ranking exhaustive or neutral — Back4App is itself on the list, ranked first, with the incentives that implies. But as an anchor for a re-audit, a third-party table that includes a self-hosted option and names the sustaining-mode context is a better starting point than any single vendor's hero price.
The three that look cheapest until bandwidth bills: Render, Fly.io, Railway
These three sit at the top of most "cheap Heroku alternative" lists for good reason: their entry prices are low, their DX is respected, and their free or trial tiers get an app running before a credit card feels it. The repricing history since 2024 is what a team migrating in late 2025 or early 2026 needs to actually understand, because each platform moved the price on a different dimension.
Render — the free tier is real, the forever is not
Render's Hobby plan still offers 750 free hours per month, which is genuinely enough to run one always-on service for free (with the sleep behavior free tiers carry). That's the cheapest Heroku-like entry point on the list by headline price: $0.
The catch is twofold and both are post-2024 changes. First, the free Postgres is not permanent — it is a 30-day trial (reduced from 90), after which the managed database bills from $7/month. A Hobby-stack-with-DB therefore stops being free on day 31 whether you changed anything or not. Second, egress overage rose to $0.30/GB past the 100 GB included, one of the highest per-GB rates among the ten. For a JSON API shipping small payloads that number never matters; for the same API once it serves assets, images, or any user-uploaded content, it becomes the largest line on the invoice.
Worked for the standard workload (API + worker + Postgres):
- Free tier month: $0 for the hobby service, then $7+ for Postgres after trial — ~$7/mo effective even before egress.
- Small prod month with 100 GB egress: ~$21/mo (2 × $7 Starter + $7 Postgres, egress still inside the 100 GB included).
- Same prod with 300 GB egress: ~$81/mo ($21 base + 200 GB × $0.30 overage).
One CX22 at €4.49 handles that same workload — two containers plus a co-located Postgres — inside the 20 TB included. The free tier wins at $0, but the crossover sits at the first paid month.
Fly.io — the cheapest per-second bill with the widest metering surface
Fly.io removed its Hobby/Launch/Scale plans on October 7, 2024. Every organization created since then is pay-as-you-go from the first second: no included VMs, no bundled transfer, a 2-hour trial and then every dimension bills. The smallest always-on shared-cpu-1x (256 MB) lands around $1.94/month, and a more typical 1 GB Machine around $5.92. That headline number is genuinely the cheapest entry price among hosted options — cheaper than Render's $7 — for a single always-on service.
The metering surface is where the 2026 story lives. Four expansions since late 2025 matter for a migrating team:
- Volume snapshots billed since January 1, 2026 — previously unmetered.
- App-scoped egress IPs at $3.60/mo each — any app that needs a stable outbound IP for an allowlist.
- Managed Postgres inter-region private networking billed since February 2026 — cross-region replication traffic that was previously bundled.
- Egress at $0.02/GB NA/EU vs $0.12/GB Africa/India — a 6× regional multiplier.
Worked for the standard workload (2 services at 1 GB-equivalent + Postgres + 200 GB egress + one egress IP + snapshots):
- NA/EU egress: ~$18-28/mo (compute ~$12-16 + volumes + $4 egress + $3.60 IP + snapshots).
- Same stack in Africa/India: ~$38-48/mo — identical code, same traffic shape, different invoice because of where the bytes landed.
One CX22 is still €4.49 for that same workload and traffic in either region. The meter is precise and fair — you pay for what you use — but the dimensions of "what you use" grew by three new meters in 2026 alone. A fixed box cannot add a meter; Hetzner can raise the box price (and did), but not start metering snapshot storage or inter-region private traffic as a separate line without changing the product.
Railway — the fastest DX with a usage pool instead of a plan
Railway's DX — railway up, project cloning, managed databases in one CLI — is broadly respected. Its pricing is not a plan in the Render sense but a credit pool: the trial is a one-time credit, then Hobby at $5/month and Pro at $20/month are minimums that buy matching usage credits, not caps. You spend the pool on compute, then pay egress and storage on top.
The consequence is that the "entry price" is not the bill. A 3-service hobby stack (API + worker + Postgres) that fits inside Render's free tier or a single Fly Machine will burn through the $5 Hobby credit quickly and bill beyond it, because each service draws from the same pool and Postgres is not free. The 2026 community-verified floor for a small production stack on Railway sits around ~$15-30/mo before egress, with the pool+overage model meaning the invoice moves month to month with traffic.
One CX22 is still €4.49 and does not ask which service the bytes belonged to.
The three that price differently enough to matter: Back4App, Upsun, and the non-platform
Not every entry on the list prices "a server that runs your code." Three of them price a different abstraction, and the CX22 comparison lands differently for each.
Back4App — you are buying a backend, not just compute
Back4App is a Backend-as-a-Service built on Parse: it bundles the database, API layer, real-time subscriptions, file storage, and auth into a platform fee, with pricing by requests and storage tiers. The free tier at 25K requests per month is generous for the Parse-shaped workload — CRUD, auth, file uploads — and the paid tiers from ~$5/mo (shared) keep that shape cheap.
The cross-over question is not "is Back4App cheaper than a CX22" but "is your app Parse-shaped." If it is, Back4App is cheaper than assembling the same stack on a CX22 because the CX22 still needs you to build the backend. If it isn't — custom runtime, non-Parse data model, workloads that outgrow the BaaS abstraction — the next bill is not a bigger Back4App tier but a migration off the BaaS entirely. That's a different cost curve than per-GB egress; it's a re-architecture cost. Price the shape, not just the month.
Upsun — four dimensions where Heroku had one
Upsun is Platform.sh's resource-based successor and prices across four independent dimensions: a flat per-project orchestration fee, CPU/memory metered per hour at different rates for apps versus services, and storage/backups billed per GB per environment. On top of that, you pick AWS, GCP, Azure, OVHcloud, or IBM Cloud per application — multi-cloud per app is a real feature, not a marketing line.
The consequence is that even a small multi-service app totals more than one CX22, because the dimensions are additive and the orchestration fee is a floor before any workload runs. A typical two-app-plus-Postgres staging of the standard workload prices in the ~$30-60/mo range depending on cloud and resource class, before any egress nuance.
The honest case for Upsun is not price but constraint: if a team is contractually required to avoid single-vendor or single-region hosting — regulated workloads, customer data-residency, enterprise procurement — per-app cloud choice is a capability no single-provider PaaS or single CX22 can match. For a team without that constraint, "choose your cloud per app" is insurance priced as a monthly fee against a risk that never materializes.
Coolify and the non-platform — the only rows that collapse to the box price
Coolify is the only entry that is not a hosted platform at all — it is a control plane you run on hardware you own. The CE is open source and self-hosted; Coolify Cloud at $5/mo is an optional managed wrapper, not the product. Installed on a CX22, its cost is the CX22: €4.49 for the same workload that cost $7-28 on the hosted rows, with the same 20 TB, and the same "no meter" property the ranking's hosted entries cannot offer.
AWS Elastic Beanstalk is the mirror image: no PaaS fee, but no PaaS abstraction either. You pay for the underlying EC2, RDS, and load balancer at their list rates and own the infra shape. The Beanstalk entry is the cheapest only if you size the raw resources smaller than the CX22 — and the moment you add the managed Postgres equivalent (RDS) and an ALB, the raw-IaaS floor sits in the ~$30-60/mo range before the team prices the operational cost of owning the VPC themselves.
The non-platform rows make the ranking's most useful point by being on the list at all: the relevant comparison for a post-Heroku decision is not only "which vendor is cheapest" but "do you want a vendor at all."
The Heroku tax nobody prices: sustaining mode
Heroku's February 6 announcement is carefully reassuring — pricing, billing, pipelines, and add-ons continue as before, existing Enterprise subscriptions renew as usual. No app was shut off. No sunset was named.
That is exactly why the tax is easy to miss. It is not a bill; it is a forward price that teams that stay will pay in three forms the current invoice does not show:
| Dimension | Staying on sustaining-mode Heroku | Migrating off Heroku |
|---|---|---|
| New features | Platform roadmap is stability only; no new primitives to plan against | New platform's roadmap is live — and its churn is your churn |
| Negotiation leverage | No new Enterprise contracts for new tenants means Salesforce has no incentive to improve the deal you renew against | Every alternative is incentivized to win the migration (credits, support, migration tooling) |
| Migration debt | Grows the longer you stay — more Heroku add-ons, more buildpack assumptions, more platform-API integrations to unwind later | Paid now, once, while docs and migration notes for Heroku-origin apps are fresh |
| Knowledge half-life | Heroku-specific operational knowledge depreciates; ecosystem tooling freezes with the roadmap | Destination knowledge appreciates — every month on the new platform is another month of runbook |
The honest counterweight is that migrating itself has a cost — a sprint of eng time, a database cutover, a period where two platforms bill at once — and for an app with no immediate operational problem and a renewal already in hand, deferring that cost is rational. The sustaining-mode tax is not that every team must leave today; it is that the longer they stay, the more expensive "later" gets, while "now" gets cheaper by comparison as migration notes and competitor credits compound.
For a team deciding between "stay and pay zero migration tax now" and "migrate and pay known cost now," the CX22 reframes the second option's floor: the hardware for the destination is €4.49 per box, not a per-seat per-service estimate to model. That doesn't make the decision free, but it bounds it.
When the CX22 actually wins — and when it doesn't
The ranking invites a single-winner reading — "which of the 10 is cheapest." The honest answer is that the winner depends on workload shape, and the CX22's fixed bandwidth is decisive for some shapes and irrelevant for others.
| Workload | What "typical month" looks like | Who wins and why |
|---|---|---|
| Hobby / side project — one service, sleeps, little egress | 1 service at 256-512 MB, 10-30 GB egress, Postgres optional | Hosted free tier wins. Render Hobby at $0 or Vercel Hobby at $0 beats any fixed box that bills €4.49 while asleep. Fly.io's per-second billing also wins for truly spiky hobby traffic. The CX22 is over-provisioned here. |
| Steady small-team prod — three services, 200 GB egress, 24/7 | 2 app services + Postgres, always-on, moderate traffic | CX22 wins on price; hosted wins on ops. One CX22 at €4.49 handles the whole stack; each hosted row is $18-40 for the same shape before the ops delta (you own upgrades, backups, and on-call on the CX22; the hosted platform owns them for you). |
| Bandwidth-heavy / global — user content, images, video, multi-region | 200-500 GB+ egress, Africa/India audience, asset serving | CX22 wins outright. Egress multipliers (Render $0.30, Fly.io $0.12 in India, Vercel $0.15) dominate the invoice; Hetzner's 20 TB included is a sunk cost. Back4App's bundled bandwidth helps but caps at the BaaS boundary. |
Two caveats the Hetzner row needs to stay honest about:
The box also got repriced in 2026. CX22 went from €3.79 to €4.49 on June 15 as part of Hetzner's generation-named catalog split that lifted CX/CAX ~30% and CCX/CPX 113-175% in some regions. "Flat rate" does not mean "never repriced" — it means repriced as a hardware contract at known intervals, not as a new meter appearing mid-year. A team that chose Hetzner for price stability in January still paid more in July; they just knew what the new price was without auditing a growing list of billed dimensions.
The CX22 is not zero-ops. The hosted rows include operator labor — upgrades, database backups, TLS renewal, scaling — that the CX22 row prices as the tenant's time. For a solo dev shipping nights and weekends, that labor is the real cost. For a team with 0.25 FTE of platform capacity or with a Cluster API fleet already running, it is sunk capacity with a known shape.
The pattern across all three archetypes is the same: hosted wins where the app uses almost nothing — and its cost curve steepens exactly where a small team stops being a side project.
What the ranking actually signals about self-hosting in 2026
A year ago, "self-hosted Heroku alternative" was a niche answer that showed up at the bottom of a roundup. Back4App putting Coolify on the same page as Render, Fly.io, and Vercel — and describing it without scare quotes — is the signal that the category has matured to where a third-party comparison treats self-hosting as a live option, not a hobby footnote.
That matters beyond pricing because Heroku's sustaining-mode shift did more than reprice a PaaS — it repriced trust in a platform's roadmap. Salesforce is not shuttering Heroku; it is no longer extending it. For a team choosing its next home, the relevant variable is not just dollars per gigabyte but who controls the next decision about what gets metered and what doesn't.
A Hetzner CX22 with a Cluster API fleet is one answer to that question where the next pricing decision is yours — when to rescale the box — rather than the platform's — which new dimension to meter. It is not the only good answer, and for hobby workloads it is not the cheapest. It is the only one on Back4App's list whose shape cannot grow a new billed line between signing and renewal.
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.