Brokenarxiv: Qwen3.6-35B Outputs on ArXiv-Derived Training Data

MathArena's Brokenarxiv dataset contains training data generated from past ArXiv articles, together with outputs generated by the Qwen3.6-35B language model. The dataset includes model answers to questions about the original statements in the articles. The dataset page was last updated on 2026-06-16.

Use Cases

• Training language models for scientific text comprehension based on ArXiv-derived content.
• Evaluating model performance on question-answering tasks related to mathematical statements.
• Analyzing the characteristics of model-generated outputs compared to original academic text.
• Studying potential biases or errors in language model reasoning for technical domains.

Strengths

• Data is derived from the established academic preprint repository ArXiv.
• Includes outputs from a specific, named language model (Qwen3.6-35B).
• The dataset page was updated on 2026-06-16, suggesting recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown.
• The description is partial, referencing a full description on an external page.

Provenance

Source

MathArena (https://matharena.ai/), with data generated from ArXiv articles.

Collection Method

Likely involves processing ArXiv articles and generating outputs using the Qwen3.6-35B model.

Freshness

Last updated 2026-06-16 14:18:07.