Gagarin vs Railway
Railway has four regions, replicas and billing on actual CPU use. Gagarin has one region, free traffic, and deletes that wait for a person. The figures, worked out.
Railway and Gagarin both take a container and give you back an address, a database and a bill. They make different bets about what you want next. This post lists the differences we could check on Railway's own pages, on 22 October 2026, and says where Railway is the better pick. Every figure links to the page it came from, in the list at the end.
The short version
Pick Railway if you need more than one region, more than one replica, a free plan, or builds straight from a git push. Its per-second billing also charges an idle service less than Gagarin's flat size rate does.
Pick Gagarin if your traffic is heavy, if you want the network policy and the secrets to come from one graph, or if a coding agent runs your deploys and you want deletes to wait for you.
Where Railway is the better pick today
Regions. Railway runs in four: US West (California), US East (Virginia), EU West (Amsterdam) and Southeast Asia (Singapore). Gagarin runs in one, Paris.
Replicas. Railway allows up to 6 replicas per service on Hobby and 42 on Pro. Gagarin runs one per service, so a node failure is a restart.
Billing on what you use. Railway charges per second for the CPU and memory a service uses: $0.00000772 per vCPU-second and $0.00000386 per GB-second. Gagarin charges for the size you choose, whether the service is busy or not. A service that sits idle most of the month costs less on Railway. The arithmetic is below.
A free plan. Railway has a Free plan with $1 of usage a month, and a trial with $5 of credit for 30 days and no card. Gagarin has no free tier, on purpose. A new account gets $5 and no card is asked for until it runs low.
Builds from your repository. Railway builds and deploys a service when you
push to its branch. Gagarin doesn't build images: your docker or your CI does,
and then gg ship or gg deploy runs it. We explain why in
what we will never build.
A contractual SLA. Railway's Enterprise plan includes one. Gagarin offers none today.
Where Gagarin is different
Traffic is free. Railway charges $0.05 per GB of egress. Gagarin charges nothing for traffic in either direction.
The graph is the policy. On Gagarin a service reaches only what it declares
with gg deps add, and declaring a database hands the service its credentials.
The graph you see in the console is the firewall and the secrets map at once.
Rotating a credential with gg resource rotate restarts every service that
holds it.
Deletes wait for a person. Destroying a service, taking an address off the internet, removing a route and transferring a project all stop and email the owner a button. Gagarin's API enforces this, whatever client asked.
The rest of the box. Alerts on one switch in the console, nightly backups
kept fourteen days, project memory for your agent, and gg eject to leave with
your Kubernetes objects. The full list is on
what's in the box. We have not compared each of these
against Railway's equivalents here.
Both have an MCP server
Railway has two: a local one that runs through its CLI, and a remote one at
mcp.railway.com that you sign in to with OAuth. Gagarin's is at
mcp.gagarin.cloud/mcp, also with OAuth. So an agent can drive either
platform.
The difference is in what a delete does. Railway's docs say its server marks destructive actions at the protocol level, and clients that respect those hints ask you to confirm. On Gagarin the confirmation happens on the server, by email, signed in as you. A client that ignores the hint still has to wait for you.
The same workload, priced
Railway's figures come from its per-second rates times the 2,628,000 seconds in a 730-hour month. Gagarin's come from its hourly rates times 730. Railway's Hobby plan costs $5 a month, and that $5 counts toward usage. Pro costs $20 per workspace with $20 of usage included.
| Workload | Railway | Gagarin |
|---|---|---|
| 1 vCPU and 2 GB, busy all month | $20.29 CPU + $20.29 memory = $40.58 | size m at $0.042 an hour = $30.66 |
| Averages 0.1 vCPU and 300 MB | $2.03 + $3.04 = $5.07 | size s at $0.014 an hour = $10.22 |
| 20 GB volume | $3.15 | 20 × $0.15 = $3.00 |
| 100 GB of egress | 100 × $0.05 = $5.00 | $0 |
A busy service costs less on Gagarin and a quiet one costs less on Railway. Every 100 GB of egress adds $5.00 on Railway and nothing on Gagarin. Put your own numbers in before you decide.
Who should pick which
If your users are spread across continents, if one replica is not enough, or if you want a free plan to try things on, Railway fits better today.
If you run a few busy services in Europe, serve a lot of traffic, and let an agent do the deploying, Gagarin is built for that. The quickest way to see is to paste one sentence into your coding agent and let it deploy: start at gagarin.cloud.