Metric Views
7
Cognaptus Model Evidence Monitor
Model Comparisons Monitor
A source-preserving dashboard for comparing model-specific benchmark signals with explicit input-price data, provenance, scope, and confidence.
Source Boundaries
Preserved
Price Basis
Declared per metric
Missingness
Explicit
How To Read This
This page is meant to answer a practical question:
How does model input price relate to a selected, explicitly sourced evaluation metric?
The current legacy view uses two sources:
- Arena leaderboard performance and vote data as an external benchmark signal
- Cognaptus pricing data from OpenRouter as a practical cost surface
New datasets retain their own metric definitions, scope, timestamps, price basis, sample sizes, and missingness. Arena preference, Artificial Analysis metrics, and task-level benchmarks are displayed separately rather than normalized into a universal score.
Model Metric Versus Input Price
Select one source-specific metric at a time. The chart does not combine Arena, Artificial Analysis, or task-level benchmarks into a common score.
Matched record
Pareto frontier
Low-confidence match
Missing or unplottable data
0Records with explicit missingness, or a non-positive cost that cannot appear on a logarithmic axis.
Low-confidence matches
0Visible on the chart with a dashed ring; inspect provenance before relying on the join.
Secondary Signal: Arena Chat Preference
Arena reflects comparative user preference in open-ended chat. It is retained as a separate, narrower signal and is not included in the metric selector or interpreted as general model capability.
Snapshot: 2026-08-29 20:39:59 +0800. Live Arena leaderboard refresh succeeded.