Small-HLE-Solved: Challenging Problems with Multi-Step Reasoning Paths

Small-HLE-Solved is a curated dataset of challenging problems from the Humanity's Last Exam (HLE) benchmark. Each problem is paired with a high-fidelity, multi-step reasoning path generated by an advanced teacher model. The dataset, created by Axiom-AI and last updated in June 2026, is formatted in JSON Lines for training next-generation reasoning models.

Use Cases

• Training language models on structured, multi-step reasoning based on curated problem-solution pairs.
• Benchmarking the reasoning capabilities of AI models against challenging problems.
• Developing educational tools for teaching advanced problem-solving techniques based on step-by-step solutions.

Strengths

• Contains curated, challenging problems from the established Humanity's Last Exam (HLE) benchmark.
• Each instance includes a structured, step-by-step solution generated by an advanced teacher model.
• Data is strictly formatted in JSON Lines, which is optimized for machine learning training pipelines.

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

Axiom-AI via Hugging Face.

Collection Method

Curated from the Humanity's Last Exam (HLE) benchmark and processed by an advanced teacher model.

Freshness

Last updated 2026-06-01 08:47:02; freshness should be verified.