Which of these has a real free tier and which does not?

Fly.io has none, stated explicitly in its own documentation, only a $1 trial. Railway's entry tier is a 30-day trial that becomes a $1 monthly tier. Northflank, Deno Deploy, Neon and Firebase all have standing free plans, though Firebase's allowance is measured in bandwidth and storage while the others are measured in requests, CPU time or resources.

Why is Neon listed next to application hosts?

Because most stacks need both and the database usually costs more than people expect. Neon is a managed Postgres service with branching, so it pairs with any of the application hosts here rather than replacing one.

Is a minimum spend the same as a fixed price?

No. Railway's $5 and $20 tiers are floors: use less and you still pay the floor, use more and the bill grows past it. That makes them easier to budget than pure pay-as-you-go and harder to predict than a flat subscription.

What happens when I exceed the free allowance?

It depends on the platform. Deno Deploy and Firebase bill the overage at published rates. Neon bills compute and storage as consumed. Railway and Northflank consume the credit balance and then charge usage. Fly.io has no allowance to exceed, so every hour is billable from the start.

Best Deployment and Database Hosting Platforms in 2026: 6 Tools Compared on What Free Actually Means

Pricing data as of Nov 7, 2026 - verify current rates on the official site before purchase

Six platforms, six different definitions of the word free. One measures requests, one measures CPU-hours, one measures dollars of consumed resources, one measures bandwidth, one charges a minimum whether you use it or not, and one has no free tier at all. Comparing the headline numbers is meaningless until you know which meter is running.

1. Fly.io: No Free Tier, and It Says So Out Loud

Fly.io runs full-stack apps and background workers on microVMs across many regions, billed by the second. It is the only platform here whose documentation states plainly that there is no free tier and no monthly free allowance for new organizations.

Pricing: New accounts get a trial of $1 in credit for Machines, Sprites and Managed Postgres, lasting seven days or two hours of machine runtime, whichever comes first, with no card required. After that it is pure usage: a shared-cpu-1x machine with 256 MB is about $2.19 per month, the same shape with 1 GB is $5.92. Reserved capacity drops the rate by 40 percent. Support starts at $29 per month, Managed Postgres starts at $38 per month, a HIPAA package is $99 per month.

Where it falls short: The trial is short by design - two hours of runtime will not tell you much about a real workload. And because there is no soft ceiling and no billing alert, an accidental deploy or a scale command with an extra zero simply lands on the invoice. Fly.io does waive unintended charges for support customers, which is a mitigation rather than a safeguard.

2. Railway: A Minimum Spend Dressed Up as Free

Railway is the smoothest of the six for pushing a repository and getting a URL back, and it bills only for active CPU and memory, by the second.

Pricing: The entry tier is $0 and starts with a 30-day trial carrying $5 of credits, then settles at $1 per month, capped at 1 vCPU and 0.5 GB RAM per service with 0.5 GB of volume storage. Hobby is a $5 minimum monthly spend that includes $5 of usage, up to 48 vCPU and 48 GB per service, 5 GB storage, one developer. Pro is a $20 minimum including $20 of usage, up to 1,000 vCPU, 1 TB storage, and unlimited workspace seats.

Where it falls short: Both paid tiers are minimums, not prices - if you use less you still pay the floor, and if you use more the bill simply grows. That makes Railway predictable only after you know your baseline. The free tier is also a trial that matures into a $1 tier, which is a different promise from the perpetual free plans elsewhere on this list.

3. Northflank: Twenty Dollars of Resources That Never Sleep

Northflank sits closer to Kubernetes than to a PaaS: services, jobs, databases, GPUs, autoscaling, and a bring-your-own-cloud mode where your own cluster runs the workload.

Pricing: Sandbox is free with no card: two services, two jobs, one database, and up to $20 per month of consumed resources on Northflank Cloud, plus one BYOC cluster up to 12 vCPU and 48 GB. Pay-as-you-go starts at $0.01667 per vCPU hour with unlimited users, deployments and workloads. Enterprise is custom, with platform fee, usage, support and SLA.

Where it falls short: The two strongest points cut both ways. No sleeping means the free allowance is spent continuously rather than only when traffic arrives, so a $20 cup runs down faster than an on-demand platform's would. And BYOC looks free until you remember your cloud provider is still billing you for the nodes.

4. Deno Deploy: Metered on Requests and CPU Time

Deno Deploy hosts edge functions rather than long-running servers, and its meter reflects that: inbound requests, egress bandwidth, and the hours your code is actually executing.

Pricing: Free is $0 with 1 million requests per month, 20 GiB egress, 10 hours of active CPU, 150 GiB-hours of memory, 10 GiB of revision storage, 5 custom domains and 10 apps. Pro is $20 per month with 5 million requests, 200 GiB egress, 50 CPU hours and 100 domains, with overages at $2 per million requests and $0.10 per CPU hour. Builder is $200 per month at 25 million requests and 500 CPU hours. Enterprise is custom.

Where it falls short: Idle apps shut down after roughly 20 to 30 seconds and are not billed while scaled to zero, which is excellent for spiky traffic and fatal for anything that needs a warm process. There is no place here for a background job that runs for an hour, and the free domain count caps how many products one account can host.

5. Neon: A Postgres Database, Priced With No Monthly Minimum

Neon is not an application host at all. It is serverless Postgres with branching, which makes it the right answer when the hard part of your stack is the database rather than the deploy.

Pricing: Free covers up to 100 projects, each with its own 100 CU-hours per month and 1 GB of storage. Launch is $0.106 per CU-hour plus $0.35 per GB-month, with no monthly minimum and 100 projects, and a typical bill around $15. Scale is $0.222 per CU-hour and $0.35 per GB-month across 1,000 or more projects, typically around $701 for heavy load at 100 GB.

Where it falls short: Compute scales to zero when idle, so CU-hours only burn while queries run - but storage is billed whether or not anyone touches the data. The free plan's hundred projects is generous for demos and previews and thin for production, where 1 GB of storage fills faster than the compute allowance runs out.

6. Firebase: A Free Tier That Ends at a Bandwidth Wall

Firebase is the least like the others: hosting is one component of a platform that also covers auth, a realtime database and analytics.

Pricing: The Spark plan is free and the Blaze plan is pay-as-you-go, and both carry the same no-cost allowance on App Hosting: 10 GiB of outgoing bandwidth per month and 5 GB of storage. Past that, uncached bandwidth is $0.20 per GiB and cached bandwidth $0.15 per GiB, with storage at $0.10 per GB. Cloud SQL for Postgres gets a three-month trial, capped at one instance per project.

Where it falls short: The allowance changed shape in August 2025, when outgoing bandwidth on App Hosting became metered after 10 GiB - a number that a moderately busy static site can pass in a week. Upgrading to Blaze does not raise the free allowance, it only removes the ceiling, which is exactly the point where a fixed bill turns into a variable one.

How to Choose

Ask what your load looks like before you compare prices. A workload of many small requests fits Deno Deploy's request meter. A database-heavy product with intermittent queries fits Neon's CU-hour model, because idle time costs nothing. A service that must stay warm fits Northflank's always-on sandbox better than a platform that sleeps. A team that wants a minimum it can budget for fits Railway.

The trap in all six is the same: the free number is quoted in a unit you do not control. Ten GiB of bandwidth, one million requests, a hundred CU-hours, twenty dollars of resources - each is a different wall, and the one you hit first is rarely the one you were watching.

Deploying something this week?

Every platform above is catalogued with its pricing model and category, so you can filter rather than guess.

Browse All Tools