A metadata-only release (v1.0) of 153 tasks from the Agents Last Exam (ALE)
benchmark for evaluating computer-use agents on long-horizon professional work.
Task Card Metadata
One row per task: titles, prompts, taxonomy, input-file descriptors
Open
Task Input Data
The input/ files each task… See the full description on the dataset page:
https://huggingface.co/datasets/Kk2406/Ll.
