Artifacts
What each model built on each task, shown on the held-out cases it was graded on.
Long context
Each model extended SmolLM2-1.7B from 8k to 64k tokens of context so that it answers lookup, tracking and counting questions about long documents. Hidden test questions are the ones the reward is computed on (each model's best and median attempt); public dev questions come from the generator the agents had, so they are not graded.