The short version
- LiteLLM: An open source proxy you run yourself. It speaks the OpenAI API. Free and open source.
- Portkey: A hosted AI gateway, and an open source gateway you can host yourself. Its site now calls it Prisma AIRS AI Gateway.
- OpenRouter: a hosted router that charges a fee when you buy credits, with a separate deal for your own keys.
- FUSR.ai: hosted LiteLLM with your own Anthropic, OpenAI, and xAI keys, routing with fusr/auto, and a flat plan.
Side by side
As of , from each company's own page. Fees change, so check their pricing pages before you choose.
| LiteLLM | Portkey | OpenRouter | FUSR.ai | |
|---|---|---|---|---|
| Who runs it | You do | A hosted AI gateway, and an open source gateway you can host yourself. | Hosted for you | Hosted for you, on LiteLLM. |
| What it adds to your bill | Nothing. You pay for your server and database. | Priced by recorded logs: Developer is free (10k recorded logs a month, 3 day log retention), Production is $49 a month (100k recorded logs, then $9 per extra 100k), and Enterprise is custom. | A fee when you buy credits: 5.5% on Standard and 8% on Business. Token prices are the provider's list price. Your provider bills you directly. The first $25,000 of list-price inference a month through your own keys has no OpenRouter fee; after that it is 5% of what the request would cost on OpenRouter. | Nothing on tokens. Your providers bill you directly. A flat plan: Free $0, Pro $29 a month, Team $149 a month. |
| Caching | Not covered here. See their docs. | Simple caching on all hosted plans, and semantic caching on Production and up. | Prompt caching is on Standard and up, not Free. | On every plan. A chat stays on the model whose cache is warm. |
| Caps and budgets | Virtual keys with max_budget and budget_duration, team budgets, and rate limits. Budgets need a Postgres database. | Virtual keys with budgets. | Budgets and spend controls are on Standard and up, and each API key can carry its own credit cap. | One monthly cap on Free. Daily and monthly caps on every key on Pro and Team. |
| Logs | Raw logs and spend data, set up by you. | Priced by recorded logs. 3 day retention on Developer. | Not covered here. See their docs. | Every routing pick logged with a reason. Prompts and replies are not stored. |
LiteLLM: run it yourself
LiteLLM is free, open source, and good software. You get virtual keys with budgets and rate limits at no fee, and you run the proxy and a Postgres database for the budgets. Self host it if you're happy to run it. FUSR.ai runs on LiteLLM, so moving between the two later is a base URL and a key. Hosted LiteLLM, explained.
Portkey: a gateway priced by logs
Portkey is priced by the logs it records, not by your token spend. It has a free Developer tier, an open source gateway you can host yourself, virtual keys with budgets, and semantic caching from Production up.
OpenRouter: one key for many models
OpenRouter is a hosted router with an OpenAI compatible API, and its Free plan has 25+ free models. The fee is charged on the credits you buy, and your own keys are fee free up to a monthly line. If you need models from providers FUSR.ai does not route, check its model list. More in the OpenRouter alternative guide.
Where FUSR.ai fits
FUSR.ai is for people who already pay Anthropic, OpenAI, or xAI by the token. It adds nothing to token prices, sends each request to a model that fits with fusr/auto, keeps prompt caching on, caps every key, and logs every pick with a reason. It routes only to Anthropic, OpenAI, and xAI, and it does not use chat plans like Claude Pro or ChatGPT Plus. See how an AI API router works and the plans.