sakana/fugu-max
Sakana
Consistency
Needs 20 scored questions.
How often it gives the same answer when asked again. How it's measured →
Alignment
Needs 20 scored questions.
How often its answer is the one most other models gave. How it's measured →
Hedging
Needs 20 questions.
How often it qualifies its answer or declines to give one. How it's measured →
A competitor's new model beats you on every benchmark I care about. Should I switch to them? →
switch 90%
hedge 10%
10 runs · consistency 77.8