About UseCaseRank

UseCaseRank is built and run by Bob Bullock. It ranks AI models by how they do on real tasks rather than on benchmarks: whether a run met 100% of the task's success criteria, and then what it cost, how long it took and how many times a human had to step in.

How it works

• A use case is a task with an exact prompt and success criteria. Where a task was inspired by a public post, the use case page links that post.
• A run is one attempt by one model at one use case, with tokens, cost, wall time, step-ins, turns, retries, setup and an evidence log recorded.
• Cost is computed from token counts and dated list prices, so it is an API-equivalent figure, not necessarily what was billed.
• Automated plausibility checks send suspicious runs to Under review, where they stay public but don't count toward rankings.
• The full scoring rules are on the Methodology page.

Harnesses used by published runs

A model's result depends on the agent harness it ran in, so every run records it. These are counted live from the published runs.

Reproduce a run

Every run page has a “Reproduce this run” section with the prompt, the model and harness settings and the success criteria. You can submit your own reproduction or a rerun on a different model. Independent reproductions raise a run's Trust % and mark it Verified.

Download the data

Every published run is available as CSV or JSON on the data download page, under CC BY 4.0. Reuse it freely, including commercially; just credit “UseCaseRank (usecaserank.com)”.

Benchmark your own task

Want your own workflow tested across models? Run my use case lets you request a paid benchmark run, with the same logging and scoring as every run here.

Money and conflicts

The site has no sponsors, affiliate links or paid placements, and rankings are never pay-to-play. Details are on the Disclosures page.