The best PaaS providers in 2026, compared on real prices

The best PaaS providers in 2026, ranked by fit rather than by hype. Ten platform-as-a-service providers, priced line by line. Lizard for a full stack that includes workers and a database. Render for flat, predictable tiers. Google Cloud Run for bursty HTTP that should cost nothing overnight. Northflank for the lowest published compute rate. Fly.io for putting the app near the user. Coolify if you want to own the box. And two names that used to top these lists — AWS App Runner and Heroku — both stopped taking on new work this year.

Written 20 August 2026. Every rate below comes from the provider's own price list that week.

---

What counts as a PaaS

A platform as a service takes your code or your container and runs it, so you never touch a server, a scheduler, or a load balancer. The line that matters in practice is what it will run: a request handler that wakes for a request and sleeps after, or a process that stays alive. Every disagreement in the table below traces back to that.

The neighbouring category is container as a service, which takes an image rather than source code. Most platforms here do both.

The ten, side by side

Best for a full stack: Lizard

Lizard bills only for active resources — $0.0278 per vCPU-hour and $0.0139 per GB-hour, by the second, on what a service measurably uses rather than the size you reserved. A background worker deploys exactly like a web service, and Postgres, Redis and S3-compatible storage come from the same place as the app.

Each service runs as a long-lived process in its own Firecracker micro-VM: no execution ceiling, a read-write filesystem, and a *.onlizard.com domain with TLS on the first deploy. lizard up builds from source with lizardpack, or uses your Dockerfile as it stands. lizard add postgres redis s3 provisions the three managed pieces, and a lizard secrets set DATABASE_URL='${{postgres.DATABASE_URL}}' --service api wires each one in by reference, so rotating a password never touches your code. lizard ssh --service <name> -- <cmd> runs a command inside the running VM. Every command takes --json, which matters if a coding agent is doing the deploying.

The catch: a service that is up bills every second it is up. Lizard does not scale to zero, by design.

If Lizard does not fit, the closest thing to it is Railway. It publishes the same compute rates to the cent and bills the same way. What you get instead is a project canvas and a longer template catalogue — a database, a dashboard, and a graph of your services in one browser tab.

Best flat pricing: Render

Render charges for the tier, not the usage: $7 a month for 0.5 vCPU and 512 MB, rising in clear steps, plus a workspace fee once you leave the free Hobby workspace. You always know the bill. Background workers, cron jobs, managed Postgres, and preview environments are all first-class, and render.yaml reads like a Procfile with more nouns.

The catch: free services spin down after 15 minutes and take about a minute to answer the next request, and you pay the full tier price whether the container is busy or idle.

Best when idle: Google Cloud Run

Cloud Run charges $0.000024 per vCPU-second, $0.0000025 per GiB-second, and $0.40 per million requests, with a free tier of 2 million requests a month. When nobody visits, the bill is zero. It also gives you a URL, a certificate, and autoscaling without a load balancer to own.

The catch: the Service is built around request handling. A queue consumer or a scheduler belongs on a Cloud Run worker pool, a separate resource that has been generally available since April 2026, and the database is a separate Google product on a separate bill. WebSockets are fine on a Service — Cloud Run supports them natively, with request timeouts up to 60 minutes.

Lowest bill on a real app: Lizard

Northflank publishes the cheapest rate card here — $0.01667 per vCPU-hour and $0.00833 per GB-hour, about 40% under Lizard and Railway — and its free Sandbox tier gives you two always-on services, a database, and two cron jobs for nothing. Read what the rate is charged on and the ranking flips. Northflank bills allocated capacity for every hour a service is deployed. So does Render, on the tier you picked. So does Fly.io, on the machine. You size for peak, then pay peak at four in the morning.

Lizard bills only for active resources: $0.00000772 per vCPU-second and $0.00000386 per GB-second, metered on what the VM measurably burns. Real traffic has seasons — weekday peaks, quiet nights, a launch, a long tail afterwards. A service that averages 25% of the CPU and 50% of the memory you gave it pays for 25% and 50%. That is where the cheaper rate card loses: 40% off a bill you pay around the clock is worse than full price on a quarter of it.

Bandwidth is the other half, and every plan comes with some:

Past the allowance it is $0.05 per GB, the same rate Railway charges — Railway meters it from the first gigabyte. Both platforms include usage credits in the plan price; only one of them also includes bandwidth. On compute rates that match to the cent, that allowance is where the two bills diverge.

The catch: a measured bill is harder to predict than a flat one. If you need a number for the budget before the month starts, Render's tiers are the honest answer.

Best for global placement: Fly.io

Fly.io runs your app as a machine in dozens of regions, close to the people using it, with Anycast routing and $0.02 per GB egress in North America and Europe — the cheapest bandwidth here, though it rises to $0.04 across Asia Pacific, Oceania and South America and $0.12 in Africa and India. performance-1x starts at $31 a month for a dedicated vCPU and 2 GB in Fly's cheapest US regions, and runs to $50.08 in São Paulo.

The catch: you write a fly.toml, and multi-region state is your problem to solve, not the platform's.

Best inside an existing cloud: DigitalOcean App Platform

From $5 a month per container, with managed Postgres, MySQL, and Valkey (Redis-compatible) a click away and Droplets, Spaces, and load balancers on the same bill. If your team already lives in the DigitalOcean console, this is the least friction available.

The catch: databases are billed separately — $7 a month for a development Postgres, more once you want one that survives the app — so the real entry price is higher than the headline.

Best if you want to own the box: Coolify

Coolify is an Apache 2.0 self-hosted PaaS with roughly 60,000 GitHub stars, a git-push deploy flow, and a catalogue of over 280 one-click services. Version 4.0 landed in April 2026, a read-only MCP server followed in 4.1 that May, and deployment control arrived in 4.3 in August 2026 — which makes it one of the first self-hosted PaaS an agent can drive natively. Run it on a $12 VPS and host several apps for the price of the server. Dokploy is the leaner, newer alternative, built on Docker Swarm with first-class Compose support.

The catch: you are now the person who patches the host, renews the certificates, and gets paged at 3am.

Two that stopped taking new work in 2026

AWS App Runner no longer accepts new customers as of 30 April 2026. Existing services keep running with no new features; AWS points leavers at Amazon ECS Express Mode. What that means, and where to go.

Heroku moved to a sustaining engineering model on 6 February 2026 — stability, security, reliability, and support, with no new features, and no Enterprise contracts for new customers. Existing apps are unaffected and pricing has not changed. It is still a perfectly good place to run an app today. It is a poor place to bet the next three years — the eight replacements, mapped onto dynos and add-ons.

How to choose in one pass

1. Does your app need a long-lived process of its own? A bot, a stateful daemon, something that holds work in memory between requests. If yes, cross off Vercel now, and put Cloud Run's worker pools rather than its Services on the list.
2. Does it need a database from the same vendor? If yes, that leaves Lizard, Railway, Render, Northflank, and DigitalOcean.
3. Is the traffic bursty with long quiet spells? If yes, scale-to-zero on Cloud Run beats every rate card here.
4. Is it busy most of the day? Then measured-use billing wins, and the per-second providers beat the flat tiers.
5. Is your time cheaper than $20 a month? Then self-host with Coolify and enjoy it.

Two neighbouring decisions, if this one is not quite yours: Kubernetes alternatives if you are weighing a cluster, and Replit alternatives if you are moving an app out of a browser workspace.

---

Try the shortest one

Free to start with $5 of credit and no card. Hobby is $5 a month.

Get started free → lizard.build/login

---

FAQ

What is the best PaaS provider in 2026? There is no single answer, but there is a short one. For a full stack with background workers and a database, Lizard. For flat, predictable pricing, Render. For bursty HTTP that should cost nothing when idle, Google Cloud Run. For the lowest published compute rate, Northflank. For running your own, Coolify.

What is the cheapest PaaS? By rate card, Northflank at $0.01667 per vCPU-hour — but that is charged on allocated capacity, so a service sized for peak pays peak all day. By actual bill for a service that is not maxed out, Lizard: $0.0278 per vCPU-hour and $0.0139 per GB-hour, billed only for active resources, plus 50 GB of egress on the $5 Hobby plan and 300 GB on the $29 Pro plan. Railway matches the compute rates to the cent but bills egress from the first gigabyte. By absolute cost, self-hosting Coolify on a $12 VPS is cheaper than all of them if your time is free.

Which PaaS providers have a free tier? Render offers free web services that spin down after 15 minutes of inactivity. Northflank's Sandbox tier includes two always-on services, one database, and two cron jobs. Railway has a Free plan with $1 of monthly credit. Google Cloud Run includes 2 million requests a month. DigitalOcean App Platform hosts three static apps free. Lizard starts free with $5 of credit for your first month and no card. Heroku has had no free tier since 2022.

Is Heroku still a good PaaS? For running an app today, yes — it is supported, patched, and priced the same as before. For planning three years out, treat it carefully: Salesforce moved Heroku to sustaining engineering on 6 February 2026, meaning no new features, and stopped offering Enterprise contracts to new customers.

Which PaaS is best for background workers? Lizard, Railway, Render, Northflank, Fly.io, and DigitalOcean App Platform all run a worker with no HTTP port exactly like a web service. Google Cloud Run runs one as a worker pool, generally available since April 2026. Vercel is the outlier: it has Queues and Cron Jobs, but no long-lived process to run a consumer in.

PaaS vs container as a service — what is the difference? A PaaS takes source code and builds the image for you. Container as a service takes an image you built. Most modern platforms do both: Lizard, Railway, Render, and Google Cloud Run will all build from source with buildpacks or accept your Dockerfile, while AWS Fargate expects an image.

Do PaaS providers lock you in? The compute rarely does — a container is a container, and every platform here will run one. The lock-in lives in the surrounding pieces: proprietary build config, platform-specific routing, managed databases you have to pg_dump out of, and add-on marketplaces. Bringing your own Dockerfile is the single most effective thing you can do to keep the exit cheap.

---

Sources

• Railway pricing, Render pricing, Fly.io pricing
• Northflank pricing
• DigitalOcean App Platform pricing, Google Cloud Run pricing
• AWS App Runner and An update on Heroku
• Coolify and Lizard pricing