Perplexity

Ranked on the overall table for the run, at per query. It returns .

Overall score

ensemble median, 0-10

Cost / query

for the whole run

p50 latency

measured, not vendor-reported

Won outright

of queries; ended in a tie

Scored cleanly

of responses

Where Perplexity holds up, and where it does not

Each bar is how much of the category-leading score Perplexity reaches.

by category

Bars start at 70%. The accent marks categories below 90% of the best available score.

A vendor that trails overall can still be the right call for one kind of query. That is why scores are reported per category. See the full table.

A cell below the coverage floor is shown as missing rather than as a thin sample.

Score is the mean of the ensemble medians in that category. "Of best" is the share of the category-leading score.
Category	Score	Of best	Behind leader	Category leader	Scored

The same table as the results page, with Perplexity marked.

Vendor	Score	Cost / query	p50 latency	Won outright	Among best	Returns

of data. Scores carry roughly points of judge-to-judge disagreement. The judges are unaudited against humans. Known limits.

One configuration is measured. If it misrepresents Perplexity (wrong tier, depth or parameters), corrections get a changelog entry, not a silent edit. Report it, or recompute it from the export.