Arxivmath Training Outputs: LLM-Generated Solutions for Mathematical Problems

MathArena provides training data generated from past ArXiv articles, together with outputs generated by the Qwen3.6-35B language model. The dataset is derived from the MathArena/arxivmath-training repository on Hugging Face. It was last updated on June 16, 2026.

Use Cases

• Training mathematical reasoning models based on problems from ArXiv articles.
• Evaluating the quality of LLM-generated solutions for mathematical problems.
• Studying the characteristics of synthetic training data for AI in mathematics.
• Benchmarking the performance of different LLMs on a corpus of mathematical text.

Strengths

• Includes outputs generated by a specific, named model (Qwen3.6-35B).
• Source data is derived from the established academic repository ArXiv.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• A bug is noted where not all rows from the source dataset received outputs, indicating potential incompleteness.

Provenance

Source

MathArena (https://matharena.ai/)

Collection Method

Generated from past ArXiv articles, with outputs produced by Qwen3.6-35B.

Freshness

Last updated 2026-06-16 14:15:14; freshness should be verified.