EvaluationsPatronus’ SpeedrunBench asks whether an agent can improve a successful route
The September benchmark separates completion from efficiency and makes the search budget part of the result.
Patronus AI
Model-building research and analysis covering Patronus AI.
Evaluations