Get your model graded

What we run

• The six Moneda Bench benchmarks, or the ones you choose, on the public split and on the private split you never see.
• The same items, time window, grader and default settings as every other system, with no retries.
• A sample of the judge's verdicts checked by people, one of them from outside Moneda with finance credentials.

What we need from you

• An API key or endpoint for your model or product, with a spend cap you set, or open weights we can run.
• The settings you want it run with, if not the defaults.
• A contact for questions while it runs.

What you get

• Your scores on each benchmark with their confidence ranges, public and private side by side.
• Per-item results for the public split, so you can see where answers lost points.
• A listing on the leaderboards, marked "run by Moneda".

Cost

Grading is a paid service: the fee covers judging, the human review of verdicts and setting up your run. Your model's own usage runs on your key. We quote the fee per system when you ask.

How results are published

The fee pays for the run, never for the result. Every system goes through the same process, and results are published whatever the score.

Running it yourself

The public split and the grader are open: you can run them yourself and send us the run files. We regrade your answers with our own grader and rerun a random sample of items against your system; a run that passes is listed as "self-reported, verified". Unchecked runs are listed as "self-reported" and are never ranked. The private split is only ever run by us.

Ask for grading

Use the grading request form. Agents and scripts can post the same request to /api/submissions (kind model); GET the URL for the schema.