HLE-Verified: A Structured Revision Benchmark for AI Evaluation

HLE-Verified is a lightweight, evaluation-ready reformatting of the benchmark created by the Skylenage Team. The original work, 'HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam', was authored by Weiqi Zhai et al. and is associated with arXiv paper 2602.13964. The dataset was last updated on February 28, III.

Use Cases

• Benchmarking model performance on structured reasoning tasks based on the 'Humanity's Last Exam' concept.
• Evaluating AI safety and alignment capabilities using a systematically revised test set.
• Training models for verification and revision tasks based on the dataset's structured format.

Strengths

• Dataset is explicitly designed for evaluation readiness, suggesting a focus on practical benchmarking.
• Based on a peer-reviewed arXiv paper (2602.13964), indicating academic rigor.
• Last update timestamp is precisely recorded as 2026-02-28 07:47:12.

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, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

Converted from the original dataset 'skylenage/HLE-Verified' by lmms-lab.

Collection Method

Lightweight reformatting of the original benchmark into a Hugging Face-native JSONL format.

Time Range

null

Freshness

Last updated 2026-02-28 07:47:12; freshness should be verified.

Geography

null