README.md

April 15, 2026 Β· View on GitHub

LLMEval-Gaokao2024-Math: LLM Evaluation on 2024 Chinese Gaokao Mathematics

LLMEval

Note: For the Chinese version of this README, please refer to README_zh.md.

πŸ”” News

  • πŸ“Š [2024-06] Evaluation results released for the 2024 Gaokao Mathematics papers.

πŸ“š Overview

The freshly released Chinese National College Entrance Examination (Gaokao) papers possess high originality and confidentiality, making them an excellent benchmark for evaluating large language models. The FDU-NLP LLMEval team presents a series of evaluations on the 2024 Gaokao Mathematics papers.

The leaderboard is continuously updated. Results are for reference only.

Key Features

  • Zero contamination β€” brand-new exam questions that models cannot have seen during pre-training
  • Dual prompt formats β€” LaTeX and escape-character versions to test prompt sensitivity
  • Two exam papers β€” New Paper I (ζ–°I卷) and New Paper II (ζ–°II卷)
  • Standardized rubrics β€” official Gaokao scoring criteria

πŸ“‹ Evaluation Design

Each model is evaluated on both papers using two different prompt formats:

PaperFormatDescription
New Paper ILaTeXMathematical expressions in standard LaTeX
New Paper IEscapeMathematical expressions in text-based escape characters
New Paper IILaTeXMathematical expressions in standard LaTeX
New Paper IIEscapeMathematical expressions in text-based escape characters

This dual-format design reveals how sensitive models are to prompt formatting in mathematical contexts.

πŸ† Results

New Paper I (ζ–°I卷)

Results are available as ranking images in the repository:

  • LaTeX format: ζ–°I卷/latexζ΅‹θ―•/
  • Escape format: ζ–°I卷/转义符桋试/

New Paper II (ζ–°II卷)

  • LaTeX format: ζ–°II卷/latexζ΅‹θ―•/
  • Escape format: ζ–°II卷/转义符桋试/
ProjectDescriptionLink
LLMEval (AAAI 2024)Foundational evaluation methodology paperarXiv
LLMEval-Fair (ACL 2026)Robust & fair evaluation, 200K+ questionsGitHub
LLMEval-Med (EMNLP 2025)Medical LLM benchmarkGitHub
LLMEval-1Phase I: General capability evaluationGitHub
LLMEval-2Phase II: Professional domain evaluationGitHub
Official WebsiteAll projects & leaderboardllmeval.com

πŸ“ Citation

@misc{llmeval-gaokao2024-math,
  author = {LLMEval Team},
  title  = {LLMEval-Gaokao2024-Math},
  year   = {2024},
  url    = {https://github.com/llmeval/Llmeval-Gaokao2024-Math}
}

πŸ“ž Contact Us


LLMEval | Fudan University NLP Lab