Our take
Friendli is a small host for downloadable models, and its cheapest rows sit well under a dollar per million input tokens. It does not train on prompts, but its compliance and data-residency posture is unverified in our data.
Use it for cost-sensitive work on the models it carries, where the cheapest row matters more than breadth, or when a no-training commitment is the thing you need. Its seven offers cover a narrower spread of models than the count suggests, so check that the one you want is there. Skip it if you need a verified security attestation or a guaranteed European endpoint, or if you need a model outside its short list.
- Its cheapest row is well under a dollar per million input tokens, so on the models it carries it is the rate to check first.
- It states that prompts are not used for training, which removes one common objection to sending work to a hosted model.
- Seven offers, with the same model family appearing twice at the same rate, so the real spread of models is smaller than the count suggests.
- Headquarters country, European hosting, an independent security attestation and a zero-retention option are all unverified in our data, so a compliance requirement needs confirming before you commit.
Point your tools here
We have not recorded what a router needs for Friendli yet — no base URL and no docs link on file. That is our gap, not a sign Friendli has no API; its own documentation is the place to look until we close it.
Privacy & data handling
These answers cover requests sent to Friendli through OpenRouter, as OpenRouter records them, last read 4 hours ago. For Friendli's own API we hold no answer: check its terms. SOC 2 and the data processing agreement below are hand-checked, not part of that sync.