Fly.io vs Railway vs Render vs Heroku — which app hosting in 2026?
As of 2026-03-16, Fly.io, Railway, Render, and Heroku use fundamentally different billing models — per-second machine runtime, per-minute CPU+RAM execution, fixed per-instance monthly rates, and wall-clock dyno billing respectively — causing cost estimates to diverge sharply based on workload shape, traffic burstiness, and network egress patterns.
Blockers
- Lock-in via package/fly-toml
- Lock-in via package/railway-toml
- Lock-in via package/render-yaml
- Lock-in via capability/heroku-review-apps
- Lock-in via capability/heroku-pipelines
- deprecated_for_new_organizations
Who this is for
- cost-sensitive
- small-team
- low-ops
- high-scale
- microservices
Candidates
Fly.io
As of 2026-03-16, Fly.io bills Fly Machines per second while running, based on named CPU/RAM presets plus roughly $5 per GB of additional RAM per 30 days (baseline US rate; higher in regions with regional markups). Stopped machines incur $0.15/GB/month for rootfs storage. Network egress is charged at $0.02/GB in North America and Europe, $0.04/GB in Asia Pacific, Oceania, and South America, and $0.12/GB in Africa and India. Fly.io deprecated its legacy free-tier plan for new organizations on October 7, 2024; new accounts have no free allowance.
When to choose
Best for microservices + high-scale or real-time + cost-sensitive where workloads run any Docker image, need persistent processes or WebSockets, or require fine-grained global region placement. Prefer Fly.io when you can right-size machines per service, traffic patterns justify per-second billing over flat-rate reservations, and your egress is concentrated in North America or Europe where the $0.02/GB rate applies. Machine reservations offer 40% discounts for predictable baseline compute.
Tradeoffs
Per-second billing rewards sporadic or short-lived workloads; idle stopped machines only pay rootfs storage. The global region network and private networking are first-class. Tradeoff: no free tier for new organizations as of late 2024, regional egress variance ($0.02–$0.12/GB) can surprise teams with multi-region or non-US traffic, and operational complexity (flyctl, machines API, volume management) is higher than Railway or Render. Migration/lock-in: standard Docker containers — near-zero app code lock-in; fly.toml is the only proprietary config file. Postgres is open-source (fly-apps/postgres-flex) and pg_dump works normally. Volume export requires app-level data transfer.
Cautions
As of 2026-03-16, the legacy free-tier plan with included resource allowances is not available to new organizations. Dedicated IPv4 costs $2/month per address; shared IPv4 is free. Egress to Africa and India is 6x the North America rate. Stopped machine rootfs storage ($0.15/GB/month) accrues even when machines are not running. Verify region-specific pricing multipliers before estimating multi-region costs.
Railway
As of 2026-03-16, Railway charges $0.000231 per GB-minute of RAM and $0.000463 per vCPU-minute (equivalent to $10/GB/month and $20/vCPU/month) plus $0.05/GB network egress. Plans (Free: $0/month with $1 credit; Hobby: $5/month with $5 credit; Pro: $20/month with $20 credit; Enterprise: custom) function as credits toward resource usage — if usage stays under the plan fee you pay only the flat rate; overages are billed directly. New Trial accounts receive a one-time $5 grant.
When to choose
Best for small-team + cost-sensitive or low-ops where usage is light and predictable and you want a simple Git-push deployment experience without managing containers or instances directly. Prefer Railway when total monthly compute stays near or below the Hobby or Pro plan credit, team size is small, and a straightforward per-minute billing model is easier to reason about than fixed instance sizing. Well-suited for side projects and early-stage production apps.
Tradeoffs
The credit-based subscription model makes costs predictable at low usage; the Hobby plan's $5/month covers meaningful runtime for small apps. Minute-level billing and a clean UI reduce ops overhead. Tradeoff: at higher usage the per-minute rates ($10/GB/month RAM, $20/vCPU/month) can exceed Render's flat instance pricing; egress at $0.05/GB is mid-range; there is no persistent free tier beyond the one-time $5 trial grant. Migration/lock-in: Dockerfile or Nixpacks (open-source) for builds; no proprietary app SDK required. Standard Postgres with pg_dump support. railway.toml is the only proprietary config file.
Cautions
Included plan credits reset monthly and do not accumulate. Usage above the plan credit amount is billed as overages with no hard cap by default — set spend limits in the dashboard to avoid surprises. The Free plan provides only $1/month of resources, which exhausts quickly under continuous workloads. For database-heavy apps, connect to Railway databases over private networking to reduce egress charges. Post-cancellation invoices may still be issued for usage in the final billing cycle.
Render
As of 2026-03-16, Render uses fixed monthly per-instance pricing: Starter at $7/month (512 MB, 0.5 vCPU), Standard at $25/month (2 GB, 1 vCPU), Pro at $85/month (4 GB, 2 vCPU), through Pro Ultra at $450/month (32 GB, 8 vCPU). All paid tiers are billed prorated by the second. Horizontal autoscaling is available on Professional workspaces and higher. Bandwidth is included per workspace tier (Hobby: 100 GB/month; Professional: 500 GB/month; Organization: 1 TB/month). Free web services spin down after 15 minutes of inactivity and receive 750 free instance hours per calendar month.
When to choose
Best for low-ops + cost-sensitive or small-team + high-scale where predictable flat-rate monthly instance costs matter more than per-second granularity, and your team wants Blueprint-driven infrastructure-as-code without managing containers directly. Prefer Render when you need horizontal autoscaling on a budget, want included bandwidth to avoid egress line items for moderate-traffic apps, or rely on Render Preview Environments for per-PR isolation. Standard ($25/month) is a practical default for production web services.
Tradeoffs
Fixed instance pricing is simple to forecast and avoids per-second billing surprises for always-on services. Included bandwidth per workspace tier covers most small and medium apps. Second-level proration on autoscaling means scale-out events do not waste full-day charges. Tradeoff: instance sizes jump in large increments (Standard $25 to Pro $85), so right-sizing between tiers can be difficult; autoscaling requires Professional workspace or higher; services with persistent disks cannot use horizontal autoscaling. Migration/lock-in: Docker or native runtimes; standard Postgres with pg_dump support. render.yaml (Blueprint) is the only proprietary config file. Persistent disks block horizontal autoscaling, which can constrain ejection timing.
Cautions
Free instances spin down after 15 minutes of idle time and take approximately one minute to restart — do not use them for latency-sensitive production workloads. Autoscaling is unavailable for services with persistent disks attached. Bandwidth overage pricing is not publicly itemized on the pricing page as of 2026-03-16; check billing docs before running high-egress workloads. Render PostgreSQL free tier expires after 30 days.
Heroku
As of 2026-03-16, Heroku bills Common Runtime dynos on wall-clock time prorated to the second. Eco dynos cost $5/month flat for 1,000 shared dyno-hours across an account and sleep after inactivity. Basic dynos cost $7/month and are always-on with 0.5 GB RAM. Standard-1X is $25/month (0.5 GB), Standard-2X is $50/month (1 GB), Performance-M is $250/month (2.5 GB, 100% dedicated CPU), and Performance-L is $500/month (14 GB, 100% dedicated CPU).
When to choose
Best for enterprise + low-ops or small-team + compliance where teams are already standardized on Heroku Pipelines, Review Apps, or Heroku add-ons, and migration cost outweighs potential savings on newer platforms. Prefer Heroku when you need HIPAA-compliant Shield dynos, a mature add-on ecosystem, or well-understood wall-clock billing for cost forecasting. Basic ($7/month) is the lowest-cost always-on option for simple production apps already on Heroku.
Tradeoffs
Heroku's billing model is mature and straightforward: per-dyno wall-clock pricing with per-second proration. The Pipelines, Review Apps, and add-on ecosystem are battle-tested. Tradeoff: Standard and Performance dyno prices are significantly higher than Render or Railway equivalents at the same RAM tier; Eco dynos sleep and share a limited monthly hour pool; Basic and Standard dynos allow only one dyno per process type, blocking horizontal scaling. Migration/lock-in: standard container support (heroku.yml or Dockerfile); Postgres export via pg_dump. Procfile is a simple, portable concept. Heroku-specific features (Review Apps, Pipelines) have no direct equivalents elsewhere and require workflow changes on ejection.
Cautions
Eco dynos sleep after inactivity, causing cold-start latency — unsuitable for latency-sensitive production APIs. Basic and Standard dyno tiers are limited to one dyno per process type, blocking horizontal scaling. Performance dynos require an established payment history before they can be provisioned. Eco plan is not available for Heroku Enterprise customers. Wall-clock billing means an idle dyno scaled above zero accrues charges continuously; scale to zero explicitly when workloads are idle.
Try with your AI agent
$ npm install -g pocketlantern $ pocketlantern init # Restart Claude Code, Cursor, or your MCP client, then ask: # "Fly.io vs Railway vs Render vs Heroku — which app hosting in 2026?"