OpenRouter vs LiteLLM: which AI gateway fits your stack?

Choose OpenRouter when you want zero infrastructure: one API key, 500+ models from 80+ providers, and you pay prepaid credits with no per-token markup (a 5.5% platform fee applies to credit purchases, $0.80 minimum). Choose LiteLLM when you want a free, self-hosted gateway with virtual keys, budgets, rate limits, and fallbacks under your own control — at the cost of running and maintaining it yourself.

The decision is mostly infrastructure vs. fees: OpenRouter trades a small platform fee for hosted reliability; LiteLLM trades your ops time for zero per-token fees.

OpenRouter: pay-as-you-go with prepaid USD credits, $0.80 minimum purchase. Provider prices are passed through at 0% markup; the platform charges a 5.5% fee on credit top-ups. A free tier gives 25+ free models from 4 free providers with a 50 requests/day rate limit. Enterprise and BYOK plans offer a fee discount after a free allowance.

LiteLLM: the open-source gateway is free forever to self-host — 100+ providers behind one OpenAI-compatible API, virtual keys, users and teams, spend tracking, budgets, rate limits, LLM fallbacks, request logging, and Prometheus metrics. Enterprise is priced annually by gateway request capacity (never per token) and adds SSO/SCIM, audit logs, secret managers, and 24/7 support; a 30-day trial key is available without a credit card.

Neither tool charges per-token markup on the model prices themselves — the difference is the 5.5% OpenRouter credit fee versus LiteLLM's self-hosting infrastructure cost (typically PostgreSQL plus optional Redis).

Deploying LiteLLM on a 24/7 VPS: Self-hosting LiteLLM via Docker on an entry-level Linux VPS (such as Hostinger VPS from $5/month) provides persistent uptime, private PostgreSQL spend tracking, and static IP whitelisting for your development team.

OpenRouter is instant: sign up, buy credits, point any OpenAI-compatible client at the hosted endpoint. There is no server to run, no database to maintain, and fallback routing is built in. It suits teams that want one billable key and no ops burden.

LiteLLM requires a self-hosted proxy (Docker or pip), a database for spend tracking, and your own deployment discipline. In exchange you get virtual keys and per-key budgets, team-based access control, and full request/response logging inside your own infrastructure — features that matter for compliance and cost governance.

Both expose an OpenAI-compatible API, so the same client tools (Claude Code, Cursor, custom apps) can switch between them by changing one base URL.

OpenRouter: you depend on a third-party service for uptime, and the 5.5% credit fee plus prepaid balance model means cash is locked in advance. Free-tier models have a 50 requests/day cap.

LiteLLM: self-hosting means you own uptime, upgrades, key security, and the database; there is no fully managed free option. Enterprise governance features (SSO, audit logs, SLAs) require the paid edition.

For a third comparison point, OmniRoute is a free open-source gateway with a strong free-provider focus — see the OmniRoute vs OpenRouter comparison in this directory.