Glamdring

Model infrastructure

LiteLLM vs OpenRouter: Fees, Hosting, Models (Sep 2026)

Choose OpenRouter for one hosted endpoint and one bill at a 5.5% fee on credit purchases; choose LiteLLM when traffic must stay in your own infrastructure or when 5.5% of monthly spend exceeds the cost of hosting and running a gateway.

Glamdring Research10 min8 sources checked

Engraving of a railway switching yard from above, with points, signals and tracks branching to engine sheds

Pick OpenRouter if you want one hosted endpoint and one bill, and can carry a 5.5% fee on every credit purchase.1 Pick LiteLLM if the gateway has to run inside your own infrastructure. Its open-source gateway is listed at $0, and its enterprise licence is priced on request.2,3 The cost of LiteLLM moves from a percentage of spend to servers and engineering time. All figures were checked 28 September 2026.

TL;DR

  • OpenRouter’s fee grows with spend. Standard credit purchases carry 5.5% with a $0.80 minimum, the Business plan lists 8%, and inference itself passes through at provider prices.4,1 At $10,000 a month of Standard credits, the fee is $550.
  • LiteLLM’s price doesn’t grow with spend. The MIT-licensed gateway is free, and Enterprise adds SSO, audit logs and support SLAs at an unpublished price.2
  • LiteLLM is built to self-host, including air-gapped. OpenRouter is a managed service with no self-hosting option.3,5
  • The coverage counts aren’t like for like. LiteLLM reports 140+ provider integrations that you connect with your own accounts; OpenRouter reports 500+ models from 80+ providers served under one account.2,4
  • The two can be stacked: LiteLLM can send traffic to OpenRouter as one of its providers.6
  • The decision turns on one comparison: 5.5% of your monthly model spend against what it costs you to host and staff a gateway.

How this comparison was checked

We read five company pages on 28 September 2026, following our research standard: LiteLLM’s homepage, its enterprise page and its GitHub repository, plus OpenRouter’s pricing page and documentation FAQ. Every fee, count and limit below is quoted from those pages. They’re company statements, so provider and model counts are reported by each company, not measured by us.

Prices are in US dollars. OpenRouter states that its credit system uses US dollars as the base currency.1 Neither pricing table sets out tax treatment, so confirm VAT or GST at checkout. Two third-party comparisons, from the gateway companies TrueFoundry and Merge, were used only for how each product is deployed and billed.

LiteLLM vs OpenRouter at a glance

DimensionLiteLLMOpenRouter
Entry price$0 open-source gateway, “Free forever”2Free plan: no platform fee, 25+ free models, 50 requests a day4
Paid tierEnterprise, price on request, 30-day trial key3Standard 5.5% platform fee; Business 8%; Enterprise fee discounts on request4
What you pay for inferenceModel providers, billed directly on your own keys7Provider prices passed through without markup, drawn from OpenRouter credits1
Own provider keysThe default way it works7No fee on the first $25,000 of list-price inference a month (Standard, Business) or $200,000 (Enterprise), then 5%4
Where it runsYour infrastructure: Docker, Helm, Terraform, air-gapped2OpenRouter’s managed service5
Reported coverage140+ providers, 1,892 unique models2500+ models, 80+ providers on paid plans4
SSOEnterprise2Enterprise only4
Contractual SLAsEnterprise, 24/7 support with SLAs3Enterprise only4
Licence or termsMIT for the gateway; enterprise features under the LiteLLM Commercial License2,8DPA via Terms of Service on Free and Standard; signed copy on request for Business4

What does each one charge?

OpenRouter charges a percentage on the money you load into it. LiteLLM charges nothing per request and sells an enterprise licence instead.

OpenRouter’s FAQ sets the fee at 5.5% of each credit purchase, with a $0.80 minimum, and says it passes provider pricing through without markup.1 Crypto payments carry 5%.1 The minimum matters for small top-ups: $0.80 is 5.5% of about $14.55, so any purchase below that pays the flat $0.80. A $1,000 card purchase carries a $55 fee, or $50 in crypto.

The pricing table lists 8% as the Business platform fee.4 On the same $1,000 basis that would be $80, but the table doesn’t spell out what the Business fee is charged on, so confirm it before upgrading. Our OpenRouter pricing breakdown goes through the plan tiers in more detail.

Bring-your-own-key traffic is priced separately. Standard and Business accounts get $25,000 of list-price inference a month with no fee, and Enterprise gets $200,000.4 Above the allowance, OpenRouter charges 5% of what the same model and provider would cost on OpenRouter.1 A Standard account running $40,000 of list-price inference on its own keys in one month would pay 5% on the $15,000 excess: $750.

Four other terms change the real cost. Platform fees are non-refundable, and unused credits can only be refunded within 24 hours of purchase.1 OpenRouter reserves the right to expire unused credits after one year.1 It doesn’t currently offer volume discounts.1 Opting in to prompt and completion logging earns a 1% discount on usage costs.1

LiteLLM’s open-source plan is listed at $0 and “Free forever”, with no credit card.2 It includes the 140+ provider integrations, virtual keys, budgets and teams, load balancing with RPM and TPM limits, and guardrails.2 Enterprise has no published price: the page offers a 30-day trial key and “Custom pricing” through a sales form.3 Enterprise adds SSO with SCIM, audit logs on every request, air-gapped deployment, a multi-region control plane and 24/7 support with SLAs.3 Its Sev 0 response target, for a full production outage, is one hour.3 LiteLLM says Enterprise sits on top of the same MIT-licensed core rather than replacing it.2

The larger LiteLLM cost is the one no page prices. The gateway runs on your own Postgres and Redis and scales with Kubernetes autoscaling.2 TrueFoundry’s comparison lists deployment, scaling, monitoring, security configuration and proxy management as work the operating team takes on.5 Put a number on that before comparing it with OpenRouter’s percentage.

Where does each one run?

LiteLLM runs wherever you deploy it. OpenRouter runs only as OpenRouter’s managed service.

LiteLLM ships official Docker images, a Helm chart and a Terraform module, and supports one-click deployment into AWS, GCP or Azure.2 It can run fully air-gapped.2 LiteLLM says the self-hosted gateway runs with no telemetry.2 Its GitHub repository also links to a hosted proxy option; the pages we checked don’t publish its terms.8

OpenRouter describes itself as a proxy that forwards your requests to the model provider.1 It logs request metadata such as timestamps, model and token counts, and doesn’t log prompts or completions unless you opt in.1 When a provider returns an error, it falls back to the next provider automatically.1 EU and US in-region routing are available on Business and Enterprise, not Standard.4

The practical consequence is the network path. Merge’s comparison puts it plainly: with LiteLLM, traffic can stay inside your own network boundary; with OpenRouter, requests pass through OpenRouter-managed infrastructure.7 If a security review requires the first, OpenRouter is out regardless of price.

Which covers more models and providers?

Both report broad coverage, but the two counts measure different things.

LiteLLM’s homepage reports 140+ LLM providers and 1,892 unique models.2 Its GitHub README describes the gateway as reaching 100+ LLM providers, so LiteLLM’s own pages don’t agree on a single figure.8 Coverage here means an integration. You still hold an account and a key with each provider. The provider list includes self-hosted inference engines such as vLLM, Ollama and Triton, and LiteLLM lets you put your own fine-tuned and self-hosted models behind the same key.8,2

OpenRouter reports 500+ models from 80+ providers on its paid plans, and 25+ free models from 4 providers on the Free plan.4 Coverage here means access under one account and one credit balance, with no separate provider contracts. Routing suffixes let you sort providers for a model by price (:floor) or by throughput (:nitro).1

The two also stack. LiteLLM supports OpenRouter’s text, chat, vision and embedding models through an openrouter/ model prefix.6 A team can run LiteLLM for budgets and access control and send some traffic on to OpenRouter’s catalogue. That traffic still pays OpenRouter’s fee.

Which should you choose?

Choose by the unit your costs grow in: OpenRouter’s cost is a share of spend, while LiteLLM’s is infrastructure and staff time.

OpenRouter fits teams that want many models without contracts, lack platform engineers, or spend little enough that the fee stays small. On Standard card purchases, $10,000 a month of credits costs $550 in fees, and $100,000 costs $5,500. In return you get one bill, automatic fallback and 500+ models with no servers to run.4,1 Teams that already hold provider keys can use OpenRouter’s bring-your-own-key route free up to $25,000 of list-price inference a month.4

LiteLLM fits three situations. Traffic has to stay inside your own infrastructure or run air-gapped. You already hold provider contracts, such as Azure, Bedrock or Vertex AI, and want to keep billing through them. Or you need hard budgets per key, team and model, with spend tracked across those contracts.2,7 Enterprise becomes a factor once SSO, audit logs or a support SLA are requirements.3

Use last month’s model bill to test the choice. Multiply it by 5.5%, then compare that figure with a quote for hosting LiteLLM and the hours needed to operate, patch and upgrade it. For the token prices underneath either gateway, see what controls AI inference cost, and for more research in this field, browse model infrastructure.

What this comparison cannot tell you

  • Enterprise prices. LiteLLM Enterprise pricing and OpenRouter’s Enterprise fee discounts are both on request.3,4
  • Your LiteLLM hosting bill. It depends on your cloud, traffic and staffing, and no page we checked prices it.
  • Latency. LiteLLM publishes its own benchmark claiming 0.66 ms of added p99 latency for its Rust gateway, measured against a mock upstream.2 That’s a company benchmark, and we have no comparable figure for OpenRouter.
  • Future fees. The view changes if LiteLLM publishes an Enterprise price, or if OpenRouter changes its 5.5% fee or its $25,000 bring-your-own-key allowance.

Frequently asked questions

Does OpenRouter mark up model prices?

No, by OpenRouter’s own account. It says it passes provider pricing through without markup and earns its money from the 5.5% fee on credit purchases instead.1 Bring-your-own-key usage above the monthly allowance carries a separate 5% fee.1

Which LiteLLM features need the Enterprise licence?

SSO with SCIM, JWT auth, audit logs, air-gapped and multi-region deployment, and 24/7 support with SLAs sit in Enterprise.2,3 Virtual keys, budgets, teams, load balancing, rate limits and guardrails are in the free open-source gateway.2

Does OpenRouter have a free tier?

Yes. The Free plan has no platform fee and offers 25+ free models, limited to 50 requests a day.4 Accounts that have bought at least 10 credits get 1,000 free-model requests a day.1 OpenRouter says free models are usually not suitable for production use.1

Are OpenRouter credits refundable?

Only within 24 hours of purchase, and only the unused credit amount. Platform fees aren’t refunded, and crypto payments are never refundable.1

Does OpenRouter store my prompts?

Not by default. OpenRouter logs request metadata, such as timestamps, model and token counts, but not prompts or completions unless you opt in for a 1% discount.1 It won’t route to providers that log prompts, or whose policy it can’t confirm, unless you switch on the model-training setting.1

Sources checked

  1. OpenRouterChecked September 28, 2026
  2. LiteLLM (Berrie AI Incorporated)Checked September 28, 2026
  3. LiteLLM (Berrie AI Incorporated)Checked September 28, 2026
  4. OpenRouterChecked September 28, 2026
  5. LiteLLM documentationChecked September 28, 2026
  6. BerriAI on GitHubChecked September 28, 2026

Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.