What the models judge
text or JSON · each question returns a typed answer
Questions
Each question comes back as a typed answer with its probability or confidence. Every column gets the same questions in its own model's format; a question a model cannot take is left out of that column and says so, and a chat model gets them in words. Run 10 times runs every column ten times and shows how often each model gives the same answer, with its averages. Each run is billed to your account at catalog prices, less the current decisions discount when there is one.
Ask anything
Pick a model — or up to three side by side — and watch each reply stream in with its cost, provider and latency. Same routing, same prices, same ledger as the API.