Request access to the ALE task data archive
This dataset bundles each task's input, software, and reference (ground-truth) data into a single archive for convenient one-shot download — in particular for running ALE locally with the local Docker provider. Because it includes the reference outputs used to score runs, access is gated to protect benchmark integrity. Approval is granted manually.
By requesting access you agree to ALL of the following. Misuse may result in access revocation and exclusion from the benchmark leaderboard.
Log in or Sign Up to review the conditions and access this dataset content.
Agents Last Exam v1.1 — Complete Task Data
122 active tasks. Use matching v1.1 task code, environment images, and data revisions.
ale-tasks-data.tar.gz contains selected inputs, setup software, and evaluator references, preserving native filenames, file modes, and symlinks. Access requires manual approval. Do not use references for model training, expose them during evaluation, or redistribute them. Stage references only after the solver stops.
Archive: 39155440558 bytes; SHA256 a7fd6eafbffcebf393580c45ce7afa4908dcd1e2ceaabcebbc867e28c7cea0a2.
Logical data manifest SHA256: def33fda086c5f1ce0064c54294af4aac92a619aeefdbed863bf230b01aa0c63.
Source code · Benchmark website
The guest images also retain inactive task payloads. Task setup replaces the butterfly animation task specification with the current software-neutral version before the agent starts.
- Downloads last month
- 3,544