MODEL & AGENT INTELLIGENCE
Performance, with a price.
Find the trade-offs. Inspect the evidence. Compare like with like.
01 / MEASURED TRADE-OFFS
The efficient frontier
Measured cohorts are only needed when an axis uses benchmark score or measured task cost.
Loading the current catalog view.
Choose axes to compare values from the same record.
02 / UNDERLYING EVIDENCE
Benchmark observations —
Select a name or plot point for provenance. Click any column heading to sort the list; missing values are never zero.
03 / SEPARATE FROM THE FRONTIER
Catalog & pricing —
Records without a measured score. Published token prices are shown independently and never used to invent task costs. Search, kind, provider, source and benchmark filters apply; the cohort selector does not.
04 / DATA HEALTH
Behind every observation, a source.
Freshness is time since collection or last successful collection, not publication. “Recent” means within 72 hours; it does not guarantee accuracy or availability.
READ BEFORE YOU RANK
Comparable is a constraint,
not a suggestion.
The chart minimizes measured USD per task and maximizes score. Use it only for metrics where higher scores are better. The API must supply the same benchmark, cohort and score unit; the cohort is the source’s assertion of a shared protocol, not independent verification.
Model and agent results may be viewed together only inside that same cohort. Missing costs, missing scores and incomplete cohort metadata stay off the chart. Price-only entries stay in the catalog. Changes to filters can change which points are efficient.
Observations are not recommendations. Hardware, harnesses, dates and source methodology can limit comparability. Follow the original evidence before making a decision.