Your requirement
API and backend hosting
Compare always-on API and worker plans first, then check free-tier sleep, timeout, network, and expiry rules as a separate operating state.
What a plan needs to be listed
Paid, public 1 vCPU and 2 GB PaaS or serverless-container records that are continuous plans or explicit always-on scenarios.
A plan reaches the table below only if all of this is true of it:
- It is a managed application platform or a serverless container plan.
- It is a paid plan, not a free tier.
- The vendor publishes a monthly price for it.
- It is exactly 1 vCPU and 2 GB, so every row on the page is the same machine.
- A public IPv4 address is included.
- It either bills continuously or is the always-on shape of the plan, because an API that sleeps is a different product.
Plans that qualify
| Provider | Plan | vCPU | RAM | Disk | Transfer | Price |
|---|---|---|---|---|---|---|
| Koyeb | always-on Checked 16 Oct 2026 1 vCPU · 2 GB · no disk 100 GB transfer, price recorded 25 Jul 2026 | 1 | 2 GB | none | 100 GB | $10.51 |
| Render | Standard Checked 18 Oct 2026 1 vCPU · 2 GB · no disk transfer unknown, price recorded 16 Oct 2026 | 1 | 2 GB | none | unknown | $25.00 |
| Render | Standard Checked 18 Oct 2026 1 vCPU · 2 GB · no disk transfer unknown, price recorded 16 Oct 2026 | 1 | 2 GB | none | unknown | $25.00 |
| Fly.io | always-on Not in the ranking: we could not find this price on Fly.io's page 1 vCPU · 2 GB · no disk 0 GB transfer, price recorded 9 Oct 2026 | 1 | 2 GB | none | 0 GB | $12.88 |
| Alibaba Cloud | always-on Not in the ranking: we could not find this price on Alibaba Cloud's page; last confirmed too long ago 1 vCPU · 2 GB · no disk 0 GB transfer, price recorded 25 Jul 2026 | 1 | 2 GB | none | 0 GB | $25.28 |
| Amazon Web Services | always-on Not in the ranking: we could not find this price on Amazon Web Services's page; last confirmed too long ago 1 vCPU · 2 GB · no disk 100 GB transfer, price recorded 25 Jul 2026 | 1 | 2 GB | none | 100 GB | $36.04 |
Plans that just miss
These came close enough to be worth knowing about. Some meet the broad requirement above but miss a narrower one; the last column says which. Listing them shows that a missing plan was checked, not forgotten.
| Provider | Plan | What it misses | Price |
|---|---|---|---|
| Render | Free web service | Not a paid plan, Not always on, Below minimum resources | $0.00 |
| Railway | Hobby | No fixed resource allocation, Full monthly compute price unknown | $5.00 |
| GitHub Pages | Pages on GitHub Free | Cannot run a backend | $0.00 |
What actually decides this
API hosting is not one workload. This comparison fixes 1 vCPU and 2 GB RAM, then admits paid public PaaS or container rows that are continuous or explicitly always-on. The current table ranks that cost grain directly from the bundle instead of preserving provider totals in prose.
The process type changes the decision. A small HTTP API that responds quickly can fit a request-oriented platform. A queue consumer needs a worker product or a container whose lifecycle is not tied to an inbound request. A long export, upload, WebSocket, or streaming response needs sourced timeout and reconnect behavior. The dataset has no fields for those semantics, so the page must not infer them from the PaaS or serverless label.
Cloud Run illustrates the caveat. Its request timeout defaults to five minutes and can be extended to 60 minutes. When a response misses the configured timeout, the connection closes with a 504 even though code may continue processing. That changes retry and idempotency requirements. Similar limits must be checked on the exact product being ranked, especially when a client can reconnect to a different instance.
Price the whole backend path. Add outbound traffic, managed database, storage, build minutes, log retention, private networking, secrets, support, and any minimum platform fee. For spiky traffic, compare the always-on total with a request or business-hours scenario using a stated invocation profile. For steady traffic, the simpler fixed instance can be easier to forecast. If protocol support, worker behavior, or private connectivity is unknown, return no winner for that scenario. A cheap process that cannot complete the request is not a cheaper deployment.
How to check this yourself
Every price here is read from the provider's own page and carries the date we read it, most recently 16 Oct 2026. The requirements above list every check we applied, so you can disagree with a threshold rather than guess at it. How we calculate the bill, and download the data if you would rather run your own.
Compare every provider on your own workload · All VPS plans · Report a wrong price