Find out which AI model handles your task, at what cost
Describe one real task. We run it across current models under the same public method we use for UseCaseRank.
Checkout isn't open yet. You can submit a task and approve its rubric now; payment opens once our test orders pass. Nothing is charged before then.
Your report covers success rate, cost per success, time, human step-ins and example failures. Everything is self-serve and in writing: no calls, no meetings, email only.
How it works
- Describe the task. The prompt and what "done" looks like (5–20 test cases help).
- Approve the rubric. We generate automated checks from your intake. Edit them, then approve. Nothing runs until you do.
- Pay. Secure checkout through Stripe.
- We run it. Every model gets the same prompt, inputs and rubric, with repeats so one lucky run doesn't decide the result.
- Get your report. A private link (printable to PDF). Questions are answered by email.
Tiers
| Quick: $49 | Plus: $149 | |
|---|---|---|
| Task | 1 | 1, with up to 3 variants |
| Models | up to 6 | up to 10 |
| Repeats per model | 3 | 5 |
| Scoring | automated checks you approved | automated checks you approved |
| Report | success, cost per success, time, step-ins, failure examples | everything in Quick, plus a cost-at-volume table and a failure-pattern summary (auto-written and labeled as such) |
| Turnaround | ~1 business day after payment | ~3 business days after payment |
| Questions | 1 email question | 3 email questions |
| Model cost cap | $10 | $40 |
Model costs are included at public list price, up to the cap. If a task would exceed the cap, we tell you before you pay.
Our rules
- Private by default. Your task, files and report are never published unless you opt in.
- If you choose to publish, the run is labeled "Commissioned", uses the same public methodology, and is listed on disclosures.
- You can't edit or veto the results. The report shows what happened.
- Commissioned runs never count in the public rankings unless an independent party reproduces them.
- Nobody can buy a ranking. Paying doesn't change scores, weights or which runs appear on UseCaseRank.
- Synthetic or anonymized data only. Don't send personal, confidential or regulated data.
- Finance tasks are synthetic test cases, not financial advice. Marine tasks are synthetic data, not for navigation.
Refunds and data
- If we can't run your task (it fails our screen or can't be graded automatically), you aren't charged.
- If a paid run fails on our side, you get a full refund automatically.
- Runs are offline and sandboxed: no network beyond model APIs, a hard spend cap and a time limit. Outputs are scrubbed for secrets and personal data before the report is built.
FAQ
- Can I talk to someone?
- All support is by email so everything is in writing.
- Which models?
- Pick from the current list below; defaults are preselected.
- Can I rerun it later?
- Yes, submit the task again to compare against new models.