Compare the recorded setting
Some sources publish several scores for a model at different reasoning-effort settings. Examenos records the effort named by the source, such as low, high, xhigh or max. Unknown means the setting has not been established for that record; it is not a default effort inferred by Examenos.
Headline and preferred effort are different
Each model/metric has one recorded headline. The interactive Prefer effort control selects a matching effort when available. If the requested setting is missing, the tool falls back to the headline and marks the fallback with an asterisk. Read that label instead of assuming every visible score used the requested effort.
Public profile and benchmark pages show headline results and retain all alternative records in expandable evidence. The interactive dashboard can apply your browser's effort preference.
Do not fill in an effort curve
A source may publish only one setting. Scores between published settings are not estimated, and a setting name is not a guarantee of identical compute budgets across providers. Inspect the source's explanation and any reported harness or cost.
Inspect a published matrix
Open recorded ARC Prize ARC-AGI-2 results and expand a model's evidence to see available efforts and source links. ARC Prize results are the primary evaluator source; vendor ARC figures remain separate.
On the Compare page, use the same effort preference for the selected models, inspect fallback labels and avoid treating a missing effort as a score of zero. See headline-selection methodology.
Sources and further reading
These explanations describe Examenos's data and display policy. Numeric examples are available in the linked model/benchmark records, each with original citations. Examenos curates results rather than conducting these evaluations.