The LiteLLM proxy is open-source (free to self-host). You still pay your underlying model providers, and there is a paid enterprise tier for SSO, audit and support. You operate and scale the proxy yourself.

LiteLLM vs OpenRouter: which AI gateway should you use?

Q: Does OpenRouter add a markup?

OpenRouter passes through provider pricing and charges a small fee on credit purchases (roughly 5%); you can also bring your own provider keys (BYOK) to avoid the per-token margin. Check OpenRouter's current pricing page for exact numbers.

Q: Can LiteLLM use OpenRouter?

Yes. OpenRouter is one of the 100+ providers LiteLLM can route to, so you can self-host LiteLLM for governance and use OpenRouter as a fallback or breadth provider behind it. They are not mutually exclusive.

Q: Which is better for production?

Neither is universally better. Choose LiteLLM when you need control, no per-token markup, your data to stay in your infrastructure, and team-level virtual keys/budgets — and you can run a service. Choose OpenRouter when you want the fastest access to many models with zero operational burden.

Q: Is OpenRouter cheaper than LiteLLM?

It depends on volume. OpenRouter has no infrastructure cost but adds a credit fee; LiteLLM has no software cost but you pay to run and scale it. At low volume OpenRouter is usually cheaper all-in; at high, steady volume self-hosting LiteLLM on your own keys often wins. Model choice dominates both — the price spread across models is far larger than any gateway fee.

The two most common starting points represent two different models: LiteLLM is a self-hosted open-source proxy you run; OpenRouter is a hosted aggregator you sign up for. The right answer depends on how much control and how little operational burden you want.

Independent · no vendor money · part of the Awesome AI Gateway benchmark · figures change — verify against each vendor's current pricing.

TL;DR

	LiteLLM	OpenRouter
What it is	Open-source proxy you self-host	Hosted aggregator (SaaS)
Setup	Deploy & operate it yourself	Sign up, one API key, minutes
Reach	100+ providers via your keys	Hundreds of models on their accounts
Cost model	Free software + your provider costs + your hosting	Provider price + small credit fee (~5%), or BYOK
Keys & billing	Your provider keys; virtual keys & budgets for your team	One key, one bill; no per-provider accounts
Ops burden	You run it (HA, upgrades, scaling)	Zero — they operate it
Data path	Stays in your infrastructure	Routes through OpenRouter
Best for	Control, governance, no markup, on your infra	Fastest access to many models, zero ops

When to pick each

Pick LiteLLM if you want your traffic and keys to stay in your own infrastructure, no per-token markup, central virtual keys + budgets + rate limits for a team, and you're comfortable running a service (Helm/Docker, HA, upgrades). It's the default open-source proxy for a reason: broad provider coverage and governance features.

Pick OpenRouter if you want one key and one bill across many models in minutes, no provider-account sprawl, and zero operational burden — ideal for experimentation, prototyping, and teams that don't want to run infrastructure. You trade a small fee and an extra hop in the data path for that convenience.

Cost: the part people get wrong

The gateway fee is usually the smaller variable. OpenRouter adds a small credit fee (or 0% with BYOK); LiteLLM's software is free but you pay to host and scale it. At low volume, OpenRouter is typically cheaper all-in; at high, steady volume, self-hosting on your own provider keys often wins.

⚖️ The dominant cost lever is which model you pick, not which gateway. The same task can vary 100×+ across models — far more than any gateway fee. Size your real workload first → cost calculator.

You can use both

This isn't either/or. OpenRouter is one of the 100+ providers LiteLLM can route to — so a common production setup is self-host LiteLLM for governance, virtual keys and budgets, with OpenRouter behind it as a breadth/fallback provider for long-tail models. You get governance and reach.

Before you trust either in production

Any proxy or aggregator sits between you and the model — so verify it doesn't silently swap models, fake-stream, or misreport token usage. Run the canary fidelity test and match the gateway's trust tier to your data's sensitivity. (This applies to every relay, not just these two.)

FAQ

Is LiteLLM free?

The proxy is open-source and free to self-host; you still pay your model providers, and there's a paid enterprise tier (SSO, audit, support). You operate and scale it yourself.

Does OpenRouter add a markup?

It passes through provider pricing and charges a small fee on credits (~5%); BYOK avoids the per-token margin. Check OpenRouter's current pricing for exact numbers.

Can LiteLLM use OpenRouter?

Yes — OpenRouter is a supported provider in LiteLLM, so you can put LiteLLM in front for governance and use OpenRouter for breadth/fallback behind it.

Which is better for production?

Neither universally. LiteLLM for control + no markup + data-in-your-infra + team budgets (if you can run a service); OpenRouter for fastest reach with zero ops.

Is OpenRouter cheaper than LiteLLM?

Depends on volume — low volume favors OpenRouter (no infra cost), high steady volume favors self-hosted LiteLLM on your own keys. Model choice dominates both.