RESULTS / 17 ARCS · SAME TEST INPUTS
Compare behavior.
Keep the differences visible.
Five measured configurations. Endpoint, clarity and ordering diagnostics are shown separately; this pilot does not assign an overall intelligence score.
Loading model results…
What you can conclude
The smallest tested model has more ordinary-endpoint clarity failures in this batch. But bigger is not better on every measure. These comparisons vary model family, precision and sometimes training—not just parameter count.
The clarity-dip check is an exploratory audit flag. A flat curve can also expose a poorly chosen arc or an intentionally unambiguous progression. Inspect the language before treating the flag as a model error.
Read all definitions and limitations →