OpenRouter vs LiteLLM: the 5.5% fee against a proxy you run

OpenRouter vs LiteLLM is a choice between renting a gateway and running one. OpenRouter is hosted: one key reaches 500+ models from 80+ providers, billed at each provider's list price, with a 5.5% fee when you buy credit. LiteLLM is an open-source proxy you host: your own provider keys at list, no fee, and a server that is yours to run (both pricing pages, read 2026-10-22).

Pick OpenRouter to call hundreds of models today with nothing to deploy. Pick LiteLLM when traffic has to stay inside your network or runs on provider contracts you already hold. The number that settles most cases is the fee: 5.5% of $1,000 of monthly usage is $55.00, so LiteLLM costs less once its server, database and upkeep cost you less than that. For the wider field, see OpenRouter and eight alternatives compared by fee.

Both put one OpenAI-compatible API in front of many providers, and both bill tokens at the provider's list price. The differences sit around the tokens: who runs the box, who holds the provider keys, where the traffic goes, and where the 5.5% ends up.

That makes the decision narrower than most feature tables suggest. If you would happily hand the gateway to someone else, OpenRouter removes the work for a percentage of spend. If the gateway is part of your security boundary, LiteLLM keeps it there and hands you the work instead.

On the same provider, OpenRouter and LiteLLM cost the same per token: that provider's list price, as Anthropic, OpenAI and the others publish it. OpenRouter adds its fee when you buy the credit that pays for those tokens. LiteLLM adds whatever its server costs you. The comparison is one subtraction, worked here at three monthly spends.

Read the LiteLLM column as a budget. At $100 a month of usage, the proxy, its database and your time have to cost under $5.50 to win. At $1,000 the budget is $55.00. At $10,000 it is $550.00, and on OpenRouter's Business plan the three budgets grow to $8.00, $80.00 and $800.00.

Two OpenRouter terms move the edges. A purchase under $14.55 pays the $0.80 minimum instead of 5.5%, so small top-ups cost proportionally more, and the fee is not refunded even when unused credit is. Unused credit can also expire a year after purchase (its FAQ, read 2026-10-22). The OpenRouter pricing explainer works the fee at every top-up size.

Upkeep belongs in the LiteLLM column too. After its October 21 release compromise, LiteLLM told affected users to rotate every provider key stored in the proxy database, the master and virtual keys, and the database password. That work is part of what self-hosting costs, and it lands on whoever runs the proxy.

• Nothing to run. Sign up, buy credit and call 500+ models from 80+ providers through one OpenAI-compatible key, with no server, database or upgrade to own (pricing page, read 2026-10-22).
• Breadth. 468 ids sat in its public model list on 2026-10-22, and a new model is one id away, with no config file to change on your side.
• Free models to try. The Free plan lists 25+ of them, at 20 requests a minute and 50 a day, rising to 1,000 a day once the account has bought $10 of credits.
• Provider failover handled for you. Fallback routing retries other providers when one is rate-limited, and an attempt that errors is not charged for its tokens.
• Open-weight models from competing hosts. OpenRouter routes models such as DeepSeek and GLM to several providers, and its model list showed some of them under the developer's own API price on 2026-10-22.
• Your own keys at volume. Up to $25,000 a month of list-price usage runs through your provider keys with no OpenRouter fee, then 5% of what the same request would cost on OpenRouter.
• Team features without hosting them. Budgets and per-key credit caps come with Standard, and Business adds 1,000 workspaces with EU or US in-region routing.

• Traffic stays yours. Requests go from your network to each provider, and LiteLLM's pricing FAQ says your data and your keys never leave your own infrastructure (read 2026-10-22).
• No fee on tokens. Every provider bills you at its list price through your own key, so at high spend the 5.5% a hosted gateway takes stays in your budget.
• Your contracts apply. Committed-use discounts, cloud credits or negotiated rates on your provider accounts flow straight through, because every request goes out on your keys.
• Governance in the free build. Virtual keys, users and teams, budgets, rate limits, fallbacks, request logging and Prometheus metrics all ship in the open-source tier.
• Deployment control. Pin the version you tested, read the code before you ship it, and run it air-gapped on Enterprise.
• Limits that are yours. Rate limits and budgets are set per key, user or team, rather than by a vendor's plan.

LiteLLM is popular open-source software, with 58.8k GitHub stars shown on its pricing page on 2026-10-22, and its release path was compromised twice in 2026. It published each incident on its own blog with the versions, the time window and the fix. These three are dated from those posts.

Two things in those posts matter for this comparison. LiteLLM answered each incident in public with the versions and the fix, and patched the CVE before it disclosed it. And the October advice shows where the risk sits in a self-hosted proxy: it holds every provider key, so one bad release means rotating all of them.

None of this makes LiteLLM the wrong choice. It means a self-hosted gateway needs pinned versions, an eye on advisories and a key rotation plan. The full timeline, with what a hosted key changes, is on the hosted LiteLLM alternative page.

You do not have to pick one. LiteLLM can call OpenRouter as a provider, and any OpenAI-compatible key the same way, so one proxy keeps your virtual keys and budgets while each model goes where you choose. Both entries below follow LiteLLM's OpenRouter and OpenAI-compatible provider docs (read 2026-10-22).

# config.yaml
model_list:
  # OpenRouter as a provider: LiteLLM reads OPENROUTER_API_KEY from the environment
  - model_name: sonnet-openrouter
    litellm_params:
      model: openrouter/anthropic/claude-sonnet-5.5
  # Any OpenAI-compatible key: openai/ prefix, api_base ending in /v1
  - model_name: sonnet-uplinkswitch
    litellm_params:
      model: openai/claude-sonnet-5-5
      api_base: https://api.uplinkswitch.com/v1
      api_key: sk-uplink-live-...

export OPENROUTER_API_KEY="your-openrouter-key"
litellm --config config.yaml

import openai

client = openai.OpenAI(
    api_key="sk-your-litellm-key",  # a LiteLLM virtual key, if you use them
    base_url="http://0.0.0.0:4000",  # the proxy
)
response = client.chat.completions.create(
    model="sonnet-uplinkswitch",  # or "sonnet-openrouter"
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Each entry is a model name your clients call. The OpenRouter entry spends credit that carried OpenRouter's fee when you bought it. The second entry bills a prepaid balance at the catalog rate, with no fee on the top-up. Keep the /v1 on the api_base: LiteLLM's docs note a Not Found error without it.

The same pattern works for any mix. Put the models you pay least for behind their own entries, give each team a virtual key with a budget, and let the proxy log spend in one place. The work of running the proxy stays with you either way.

Like OpenRouter, UplinkSwitch is hosted and needs no provider keys. Unlike it, prices come from its own catalog rather than the provider's list: on 2026-10-22, 58 of the 75 chat models with a list price sit under it on both input and output. The table puts the four ways of paying for the same model side by side.

On the four-model mix in the cost table, $1,000 of list-price usage costs $422.22 here, with no fee on top-ups and nothing to deploy. Credit is prepaid from $10 with no subscription, and each top-up locks your rates for 30 days. Every chat model's rate next to its list price is in the LLM API pricing comparison, and the other hosted gateways are in the gateway fee comparison.

Two checks come with it. The daily real-model check sends every model the same prompt once a day (230 of 269 stable at the last run), and every non-streamed answer carries a signed receipt you can check at statements/verify without an account. One key serves every model in the OpenAI format, the Claude models in the Anthropic format and the Gemini models in the Gemini format, at one price per model, with per-key spend limits, rate limits and model allowlists. The SDK setup guide covers each format. Behind LiteLLM, it is the second entry in the config above.

Is LiteLLM better than OpenRouter?

Neither wins across the board. LiteLLM is the better pick when the gateway has to run inside your network, on provider keys or contracts you already hold, or when 5.5% of your monthly spend is more than a server costs you. OpenRouter is the better pick when you want 500+ models on one key today, free models to try, and nothing to deploy or patch (both pricing pages, read 2026-10-22).

Is OpenRouter cheaper than LiteLLM?

On the same provider they cost the same per token: that provider's list price. OpenRouter then adds 5.5% to each credit purchase, and LiteLLM adds your server. At $1,000 of monthly usage the fee is $55.00, so LiteLLM is cheaper if its server, database and upkeep cost less than that, and OpenRouter is cheaper if they cost more. A hosted key priced under list is a third answer: on the four-model Anthropic and OpenAI mix above, $1,000 of list-price usage costs $422.22 on UplinkSwitch, under the list price both of them start from.

Can LiteLLM use OpenRouter?

Yes. LiteLLM's docs route any OpenRouter model with the openrouter/ prefix, for example openrouter/anthropic/claude-sonnet-5.5, using an OPENROUTER_API_KEY (read 2026-10-22). You then run the proxy and also pay OpenRouter's fee on the credit behind it. Any other OpenAI-compatible key works the same way, with the openai/ prefix and an api_base that ends in /v1.

Is LiteLLM free?

The open-source proxy is free forever, with no license fee, and includes virtual keys, budgets, rate limits, fallbacks and spend tracking (litellm.ai/pricing, read 2026-10-22). Enterprise adds SSO, SCIM and audit logs, priced by annual request capacity and never per token. The tokens are not free: each provider bills you at its list price, and the server is yours to pay for.

Does OpenRouter mark up model prices?

No. Its pricing page says inference is billed at the provider's list price on every plan. The fee is charged when you buy credits: 5.5% on Standard and 8% on Business, with a $0.80 minimum, and 5% on crypto (pricing page and FAQ, read 2026-10-22). Platform fees are not refunded, and unused credit can be refunded only within 24 hours of the purchase.

Portkey vs LiteLLM vs OpenRouter: which one?

Portkey is a gateway over your own provider keys, priced by a plan per logged request: Developer is free with 10,000 logs a month and Production is $49 a month with 100,000 (portkey.ai/pricing, read 2026-10-22). Palo Alto Networks closed its acquisition of Portkey on May 29, 2026 (its press release). LiteLLM is the free, self-hosted option, and OpenRouter is hosted at list plus 5.5% on credit. Pick by who should run the gateway and whether you want to hold provider keys; the Portkey alternatives page prices one month on each.

OpenRouter, LiteLLM, Portkey and Palo Alto Networks are trademarks of their owners. This page quotes OpenRouter's pricing page, FAQ and model list, LiteLLM's pricing page, provider docs and security posts, and Portkey's pricing page, all as read on 2026-10-22, and the Palo Alto Networks press release of May 29, 2026. Our figures come from GET https://api.uplinkswitch.com/v1/catalog and GET https://api.uplinkswitch.com/v1/proof.

More on each side: UplinkSwitch vs OpenRouter, one to one, Portkey alternatives priced and Requesty alternatives. Get a key: prepaid from $10, no subscription, 30-day price lock, in the console.