5.5% vs $0
OpenRouter fees the credits; LiteLLM is free software plus your ops time
Blog
OpenRouter vs LiteLLM: the honest comparison
One is a hosted catalog with a 5.5% fee; the other is free software plus your operations time. The decision is a constraint check, not a feature war, and the math has a punchline.
~$1,000/mo
what self-hosting actually costs at two loaded engineer-hours a week
1 question
can data leave your infrastructure? If no, LiteLLM wins by default
OpenRouter is a hosted aggregator: 400+ models, list prices, 5.5% fee on credits, zero ops. LiteLLM is a self-hosted gateway: free software, your infrastructure, your ops time, roughly $1,000/month at two loaded engineer-hours a week. Data locality forces LiteLLM; no ops capacity forces OpenRouter; between them, do the spend math.
The structural difference
OpenRouter runs the gateway for you and fees the money you load: 5.5% on credits, tokens at provider list prices, BYOK free to a million requests a month. LiteLLM hands you the gateway as open-source software: no fee, no markup, and every operational responsibility, upgrades, incidents, rate-limit handling, on your pager.
Both speak the OpenAI-compatible surface, so application code barely differs. What differs is who operates the layer and who chooses models: on both, that second job is yours, in a dropdown on one and a config file on the other.
The real monthly math
Self-hosting is free the way a puppy is free. At two loaded engineer-hours a week and $120 an hour, LiteLLM costs about $1,000 a month in time before the first incident. OpenRouter's fee only reaches that number around $18,000 a month of spend. If your team already runs platform infrastructure with on-call, the marginal ops cost drops toward zero and the math flips hard toward LiteLLM.
That conditional is the whole comparison: LiteLLM is the cheapest option in the category for teams with genuinely spare ops capacity, and the most expensive one for teams without it.
Every fee structure at your spend
OpenRouter 5.5% on credit top-ups; Requesty 5% token markup; Portkey $49/month flat; LiteLLM $0 licence with ops estimated at ~$1,000/month (two loaded engineer-hours a week) unless the box is ticked. Fees verified 2026-07-28. The crossover between 5.5% and $49 flat sits near $891/month.
Where each wins outright
LiteLLM wins on data locality, full stop: traffic stays in your infrastructure, which no hosted option can offer. It also wins on markup purity; nothing is feed because nothing is hosted. OpenRouter wins on time-to-first-request, catalog breadth, and everything operational; there is no version to upgrade and no 2am page.
Neither answers the selection question. If the reason you are shopping is that model choice keeps reopening every time the market reprices, that is the job a blended model removes: Pareto runs several LLMs per request and bills at cost, receipts on the model card. Different category, ours, judge accordingly.
Questions, answered
Is LiteLLM really free?
The software is. The cost is operations: hosting, upgrades, incidents, and the engineer-hours to own them, roughly $1,000/month at two hours a week of loaded time. With spare platform capacity, that marginal cost can approach zero.
Which is faster?
We measured a full routing gateway hop at +68ms median versus direct, with better p95 than direct. A well-placed self-hosted LiteLLM inside your own network can beat that; a neglected one can be far worse. Operations, not architecture, decides.
Can I migrate between them easily?
Yes. Both expose the OpenAI-compatible surface, so the move is a base URL and key change. Routing configs and provider preferences do not travel; budget an afternoon to re-express them.
Which should a startup pick?
Under ~$900/month of spend with no compliance constraints: OpenRouter, and spend your hours on product. Data-locality requirements or an existing platform team with on-call: LiteLLM. Full reasoning in our gateway guide.
Compare any two models
VS
Rates verified 2026-07-28. "Measured task" = our identical dashboard-generation prompt, metered where marked ✓ and list-math otherwise. Verbosity from the Verbosity Index, Edition 1. Data: prices.json.
Run the comparison on your own traffic. $100 to verify.
Buy Pareto API credits