RESEARCH
PROMPTS.
Which language model converts a rough idea into the strongest research brief? Comparing Skill v2.2 outputs on a shared rubric.
Provisional results · 1/3 benchmarksModel rankings
| # | Model | Research /35 | Journalism /25 | Factual control /25 | Clarity /15 | Overall | Details | Compare |
|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||
Don’t take our word for it.
01 / Same task
Every model receives the same benchmark request. We judge the resulting research prompt, not a finished history article.
02 / Four dimensions
Research design 35%, journalism 25%, factual control 25%, clarity 15%.
03 / Penalties
Deduct points for unsupported claims, invented sources, preloaded hypotheses, and output-contract failures.
04 / Provisional by design
These are editorial ratings from the first benchmark. Official standings require three matched tasks per model.
This ranking evaluates prompt quality, not historical accuracy of downstream articles or general model ability. Repeated runs are independently scored and ranked by their arithmetic mean, rounded to a whole point. Unavailable category breakdowns are shown as —; no values are inferred.