Independent testing Updated August 2026 387 self-hosting guides 5 VPS providers tested

comparison

Best VPS for Jan Server (2026): Sized for the 8 GB Floor

Jan Server needs 8 GB RAM minimum, so most cheap VPS plans miss it. Compare the plans that clear the floor and see which fits remote versus local inference.

Last verified:

Jan is an open-source ChatGPT alternative. The part you put on a VPS is Jan Server, a Go microservices platform that exposes an OpenAI-compatible API, and it is considerably heavier than the single container most hosting comparisons assume.

That matters for shopping, because the project’s README states a floor of 8 GB RAM minimum, 12 GB recommended for the full service set. A 4 GB plan at 5.49 EUR does not meet it. Before comparing providers, you need to settle one question.

The question that decides your plan size

Jan Server’s setup wizard asks which LLM provider to use, and the answer changes your hardware needs more than anything else:

Nearly everyone self-hosting Jan Server on a normal VPS is in the first category. The plans below are sized for it.

VPS plans that clear the 8 GB floor

ProviderPlanvCPU / RAM / DiskPriceBasis
ContaboCloud VPS 44 / 8 GB / 100 GB SSD5.50 EUR/moexcl. VAT
ContaboCloud VPS 66 / 12 GB / 200 GB SSD7.50 EUR/moexcl. VAT
Hetzner CloudCX334 / 8 GB / 80 GB NVMe8.49 EUR/moexcl. VAT
netcupVPS 1000 G124 / 8 GB / 256 GB NVMe10.36 EUR/moincl. 19% VAT

Check the basis column before ranking these. netcup quotes gross while Contabo and Hetzner quote net, so netcup’s 10.36 EUR is roughly 8.71 EUR net and the gap is smaller than the raw numbers suggest.

Contabo Cloud VPS 4 — the cheapest plan that qualifies

Contabo’s entry plan is 8 GB, which lands exactly on Jan Server’s documented minimum at the price other providers charge for 4 GB. The 100 GB disk absorbs Postgres growth and the container images without planning. Contabo’s provisioning is slower than Hetzner’s and there is no comparable API, so this is the pick when you are standing up one server rather than automating a fleet.

If you expect several concurrent users, Cloud VPS 6 at 7.50 EUR gives you the 12 GB the README recommends, and that is the version to buy if the budget stretches.

Hetzner Cloud CX33 — when you automate

CX33 costs about 54% more than Contabo’s 8 GB plan for the same RAM and a smaller disk. What you get back is NVMe storage, a genuinely good API with Terraform support, and faster provisioning. Note that Hetzner’s cheaper CX23 at 5.49 EUR is a 4 GB plan and sits below Jan Server’s minimum, so it is not the plan to pick here even though it appears in most Hetzner recommendations.

DigitalOcean — documentation over price

DigitalOcean’s entry droplet is 1 GB, well under what Jan Server needs, so reaching 8 GB means moving several tiers up the price list and paying noticeably more than the European providers above. The reason to accept that is the ecosystem: thorough documentation, predictable networking, and one-click images for the Postgres and Redis pieces you may want managed separately. Vultr and Linode sit in the same position, with 1 GB entry plans that need the same upgrade.

Disk, not just memory

Jan Server pulls a broad set of images: the Go services, Postgres, quay.io/keycloak/keycloak:24.0.5, and Kong. Budget 40 GB before any data, and more if you enable the Media API with local storage. If you later switch on local vLLM, model weights land in a cache volume and the requirement changes completely. janhq/Jan-v1-4B is small as models go, but a model cache is the fastest way to fill a 40 GB disk.

This is where Contabo’s 100 GB and netcup’s 256 GB earn their place over Hetzner’s 80 GB.

Which to choose

Buy Contabo Cloud VPS 4 if you want the cheapest server that honestly meets the requirement, and Cloud VPS 6 if you want the recommended 12 GB. Choose Hetzner CX33 if you provision through Terraform or want NVMe. Choose DigitalOcean if its documentation and managed add-ons save you more time than the price difference costs.

Whichever you pick, size for 8 GB rather than the 4 GB plans that dominate cheap-VPS lists. For the full lineup, see our VPS comparison.

Frequently asked questions

How much RAM does Jan Server actually need on a VPS?

The project's own README asks for 8 GB minimum and 12 GB for the full set of services. That is the floor for the Go services, Postgres, Keycloak and the Kong gateway on their own, before any model is loaded locally. The 4 GB plans most comparison articles recommend are below the documented minimum.

Can I run Jan Server on a VPS without a GPU?

Yes, if you point it at a remote OpenAI-compatible endpoint during setup. The wizard offers that as an alternative to local vLLM, and it keeps the VPS handling only the API and orchestration layers. There is also a CPU inference profile, but generation speed on shared vCPUs makes it useful mainly for testing.

Is Jan the desktop app the same thing as Jan Server?

They are two parts of the same project. The desktop app runs models on your own machine and is not what you install on a server. Jan Server is the self-hostable piece: a multi-user LLM API platform with an OpenAI-compatible interface, which is what a VPS guide is actually about.