Download the data
Every published run (not under review), one row per run. Updated within about 5 minutes of a new run being published.
- runs.csv: one row per run, token counts summed across models.
- runs.json: the same runs with a per-model breakdown (tokens, cost, settings).
License: the data is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). You can share and adapt it, including commercially, as long as you credit “UseCaseRank (usecaserank.com)” with a link to usecaserank.com and say if you changed it. Example credit: Data: UseCaseRank (usecaserank.com), CC BY 4.0
.
Not included: submitter accounts, emails, API keys, raw submissions, free-text setup notes and evidence logs (each row links its run page, where the log is shown). How the numbers are produced: Methodology.
CSV columns
- run_id
- run_url
- run_date
- · When the run happened (ISO 8601, UTC).
- published_at
- · When it was published on the site (UTC).
- use_case_slug
- use_case_title
- category
- models_label
- · Display names, with settings such as effort, joined by " + ".
- model_slugs
- · provider/model for each model, separated by ";".
- input_tokens
- · Sum over the run's models (same for the other token columns).
- output_tokens
- cached_tokens
- cache_write_tokens
- cost_usd
- · Computed cost if available, else the claimed cost (USD, list price).
- computed_cost_usd
- claimed_cost_usd
- cost_is_estimate
- · True when the cost is a list-price estimate rather than a billed amount.
- wall_time_sec
- user_interventions
- · Human step-ins after the first prompt.
- turns
- retries
- first_try_success
- outcome
- · success (100% of the rubric), partial or fail.
- autonomy
- quality_self_score
- · Optional 1–5 self-rating. Left out of the files while no published run has one.
- agent_tool
- execution
- gpu
- vram_gb
- quantization
- trust_score
- trust_status
- · ok or verified (under-review runs are not exported).
- is_example
- parent_run_id
- evidence_sha256
- · SHA-256 of the evidence log, when submitted. The log itself is on the run page.
- replaces_run_id