Agents Last Exam: Metadata for 147 Professional Task Cards

147 task cards form the metadata layer of the Agents Last Exam benchmark for evaluating computer-use agents. This version 1.0 release includes titles, prompts, taxonomy, and input-file descriptors for each task. The dataset is published by agents-last-exam on HuggingFace and was last updated on June 5, 2026.

Use Cases

• Benchmarking agent performance on long-horizon professional tasks based on the provided task prompts and taxonomy.
• Analyzing the distribution of task types and complexity based on the task card metadata.
• Filtering or selecting specific task categories for targeted agent evaluation based on the taxonomy labels.
• Preparing input data for agent simulations based on the referenced input-file descriptors.

Strengths

• Contains metadata for 147 distinct evaluation tasks.
• Provides a structured taxonomy for categorizing tasks, which suggests systematic organization.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

agents-last-exam on HuggingFace

Freshness

Last updated 2026-06-05 04:06:13; freshness should be verified.