What Examenos does
Examenos curates published benchmark results and API pricing for an owner-approved set of models. We do not conduct the evaluations displayed here. Source links are the basis for every recorded score.
Official and independent results
Official records come from vendor announcements, model cards, reports and comparison tables. Own-vendor, peer-vendor and secondary reporting are identified in evidence. Independent results name their evaluator and stay separate, even when benchmark names overlap.
Source selection and exact identity
We check benchmark primary references and model releases, including chart/image/PDF figures. Released checkpoint, revision, subset, harness and units must match the record. Unlisted peer models are not automatically added. Discovery ledgers are leads to primary evidence, not numeric sources.
One default headline, alternatives retained
Each model/metric has one headline. Own-vendor launch evidence normally wins over peer comparisons; snapshot identity, evaluation date, configuration and precision resolve other conflicts. Alternatives are retained with sources and notes. Shared application helpers pick the same default value throughout the site. Interactive effort selection falls back to the headline with a visible asterisk when the requested setting is missing.
Missing data and units
An em dash means no recorded result, not zero or failure. Coverage counts distinct model/metric pairs and describes dataset completeness, not capability. Official and evaluator denominators are separate. Published 0–1 benchmark values are stored as percent with the original scale noted; monetary evaluator metrics stay distinct from token prices.
Dates and updates
Source dates and as-of dates describe publication or the recorded check, as specified in the evidence. Timeline stops are repository recording dates displayed in Dubai time, not evaluation dates. The earliest retained snapshot is a baseline, not proof of first publication. Static public pages regenerate from committed data when the site is built. Alternative records remain reachable through evidence.
Prices and speed
Recorded token rates are USD per 1M tokens with cached dates. Published contributor tiers are preferred when present. Healthy OpenRouter endpoint minima are selected separately for input, output and cached input; different metrics may use different providers. The dashboard can enhance snapshots with live data. Published parameter counts and licenses are recorded from official sources rather than inferred.
Corrections and limitations
Feedback suggestions are untrusted submissions and reviewed before changes. Corrections require primary evidence. The default view is not a universal model ranking, and differing setups can limit comparisons. Use the Feedback channel for source-backed corrections, inspect the model profiles, or read our comparison guides.