Benchmark RadarTeXFix-Bench
TeXFix-Bench evaluates LLM-based full-source document repair across LaTeX, Typst, and Markdown. It provides 10,437 repair instances derived from 743 openly licensed seeds, with a fixed zero-shot protocol and provider-pinned routing.
- Released
- 2026-08-07
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
The benchmark fills a gap in document-repair evaluation by grounding faults in a mined taxonomy, offering a reproducible protocol to compare models on compile success and content restoration. This supports practical selection of models for document repair tasks.
Motivation
Scientific and technical writing depends on markup sources that must compile: LaTeX, Typst, and Markdown pipelines fail on missing delimiters, mismatched environments, broken imports, or package conflicts.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.