# Take the Finest Challenge

> Three models answer the same prompt. One is frontier. See if you can spot it before the prices are revealed.

Canonical page: https://finest.so/challenge

## How it works

Enter a prompt or choose a self-contained example. Finest returns three anonymous answers to the same request. One answer is from a frontier model and two are from lower-cost models. Model identities and reference prices stay sealed until the visitor votes.

The visitor can pick answer A, B or C as frontier, say the answers feel the same, or reject all three. The reveal names the models, shows their reference costs and contributes one anonymous closed verdict to a rolling seven-day tally. The public pulse reports how often players did not pick Frontier, overall and by a small task category.

## Availability

Production deployments with the live challenge configured accept text prompts and limit each IP address to three rounds per minute. Local and staging deployments use explicitly labeled recorded examples instead of calling providers. Files selected in the interface stay on the visitor's device and cannot start a live document challenge.

## What a result means

Frontier models are for frontier tasks. This is one blind preference game, not a benchmark, an overall intelligence claim or a routing recommendation. Finest only authorizes production routing from customer-specific, governed evidence that clears the customer's quality bar.

## Share a result

Every new share is an addressable round, not a generic result state. A recipient receives the exact prompt and three anonymous answers, makes a fresh choice, then sees both the models and the original player's verdict. Recorded examples use compact deterministic links such as https://finest.so/challenge?round=d1.writing.A0.B. Live rounds use an encrypted, expiring, stateless capability link; the prompt and answers are not written to a Finest database. Shared round pages remain non-indexed while the canonical page stays https://finest.so/challenge.