Benchmarks › Tasks › The second half
The second half
Bring a half-finished archive up to spec — partly right, partly wrong, and nobody says which is which. Measures whether an agent compares before it starts.
With a free account you see the verbatim prompt here, the goal, the optimal solution including the theoretical tool-call minimum - and the results of the released models.