Fly.io Usage Rates, Support Fees & Real Costs: 2026 Guide

By usage, not by plan. Fly.io charges for the micro VMs, storage and bandwidth you run, with per-resource rates set out in a calculator rather than on the plans page. That means there is no single monthly figure until you model your own workload. A small always-on app costs a few dollars; a busy one costs whatever it consumes. On top of usage, paid support starts at $29 a month and HIPAA compliance is a flat $99. Machine reservations cut 40 percent off compute for capacity you commit to in advance.

Fly.io provides a small free allowance to trial the platform rather than a standing free plan. It is enough to deploy an app and evaluate whether Fly.io suits your stack, but not to run production. Once you exceed the allowance, usage bills at the per-resource rates in the calculator. Free and unpaid accounts also get only community support, not the paid SLA. Treat the free allowance as an evaluation window, and budget from your modelled usage the moment the project becomes real rather than a test.

The ones that sit off usage. Paid support with a response SLA starts at $29 a month and is separate from compute. HIPAA compliance for regulated health data is a flat $99 a month, covering the BAAs and SOC2 controls, so a tiny workload with HIPAA on effectively starts near $99. Beyond those, everything is metered: compute, volumes and bandwidth all bill per resource. The important thing is that the plans page shows none of this as a total, so the calculator is the only honest source of a monthly figure.

Serious paid support starts at $29 a month, which adds a response SLA, technical guidance and access to Fly.io's engineers. It also carries a refund policy for accidental deployments, which free and unpaid accounts do not get. Higher support tiers cost more for faster response and deeper help. Community support remains free but comes with no guarantees. For a hobby project, community help is usually enough. For anything where uptime matters to revenue, the $29 tier or above is a reasonable line on the usage bill.

It depends on how critical the workload is. Free accounts rely on community support, which is fine for side projects and experiments where a slow answer costs nothing. Once an app earns money or serves users who expect uptime, the $29 support tier buys a response SLA and direct engineer access. That can pay for itself the first time something breaks. The refund policy for accidental deployments is a bonus at that level. Add support when reliability starts to matter, and skip it while you are still evaluating the platform.

A machine reservation cuts 40 percent off the on-demand compute rate in exchange for committing to capacity in advance. It is the largest published saving on the platform, but it only pays off on machines you run steadily, since you prepay the capacity whether or not you use it. The rule of thumb is to reserve the compute that stays on every hour and leave spiky or uncertain workloads on demand. Reserving before your usage settles risks paying for machines you later scale down, so model first, then commit.

Yes. At the Enterprise level, Fly.io quotes custom terms, which puts compute rates, reservations and support on the table together. A steady, sizeable account carries the most weight, particularly with a Railway or Render figure and a committed-use offer behind it. The closing weeks of a quarter tend to loosen a rep's number, so aim there. Self-serve usage, by contrast, does not negotiate. There the saving is reserving your baseline and modelling the workload before launch.

Model the workload in the calculator before deploying, so there are no surprises, then reserve only the compute that runs every hour for the 40 percent cut. Leave spiky or uncertain work on demand rather than locking prepaid capacity. Add paid support and HIPAA only when the project genuinely needs them, since both are flat lines on top of usage. Keep bandwidth and storage in check the same way you would on any cloud. Stacked, those moves hold a Fly.io bill close to the usage you actually consume.