Finest is the LLM gateway that charges a percentage of savings: the model you name serves by default, a cheaper one serves only where a sealed, published test proves it holds your quality bar on that exact task shape, and every request returns a receipt naming both models and both prices. Tokens at the provider's published rate with no markup; the only fee is 25% of the saving proven on a request. No saving, no fee.
Site: finest.so · Product description for machines: finest.so/llms-full.txt · Comparisons and guides: finest.so/answers.md · Research: finest.so/research
| Repository | What it is |
|---|---|
| blindtest | A blinded triangle-test between two LLMs on your own prompts. Exact statistics, tell audits, MIT. Can you actually tell them apart? |
| llm-price-history | Published LLM prices as versioned open data: every record with the provider's URL, date, and quoted sentence. History and drift as queries. |
| inference-receipts | Inference Receipts: a signed, content-free record of what actually served an LLM request. Spec, threat model, forty-line verifier. |
The common thread is the company's one idea: claims about model serving should be specific, published, and checkable by someone who does not trust the claimant.
This organization, finest.so, x.com/usefinest, and youtube.com/@Finest-AI. An account absent from this list is not Finest, whatever it is named.