Research map / Large language models / Agents, code and math
ACADREASON: Exploring the Limits of Reasoning Models with Academic Research Problems
Xin Gui, King Zhu, JinCheng Ren, Qianben Chen, Zekun Moore Wang, Yizhi LI, Xinpeng Liu, REN WENLI, Linyu Miao, Tianrui Qin, Ziqi Shu, He Zhu and 6 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| give GPT-5 all hints (background, definition, methodology) (ablation) | GPT-5 · ACADREASON | pass rate / checklist score: 16.0/40.5 -> 40.0/67.8 | Improved |
| add methodology hints rather than background hints (ablation) | GPT-5 · ACADREASON | checklist score: 42.5 (background) vs 64.3 (methodology) | Improved |
| add background hints (ablation) | GPT-5 · ACADREASON | pass rate: 16 -> 16 | No clear effect |
| use a reasoning model instead of a general model in the same family | DeepSeek-V3 vs DeepSeek-R1 · ACADREASON | checklist score: 15.9 -> 23.8 | Improved |
| use an agent framework (OAgents with GPT-5) instead of the plain LLM | GPT-5 · ACADREASON | checklist score: 40.5 -> 65.1 | Improved |
| use newer model version | GPT-4.1 to GPT-5 · ACADREASON | pass rate / checklist score: 0/21.0 -> 16/40.5 | Improved |
| use newer model version | DeepSeek-V3 to V3.1 · ACADREASON | checklist score: 15.9 -> 24.8 | Improved |
| provide hints in STEM (CS, Math) versus humanities (Econ, Law, Phi) (ablation) | all models · ACADREASON by discipline | average gain | Mixed |
Closest papers by meaning
- From Passive to Active Reasoning: Can Large Language Models Ask the Right Questions under Incomplete Information? (ICML 2025 · similarity 0.90)
- Putnam-AXIOM: A Functional & Static Benchmark for Measuring Higher Level Mathematical Reasoning in LLMs (ICML 2025 · similarity 0.90)
- UGMathBench: A Diverse and Dynamic Benchmark for Undergraduate-Level Mathematical Reasoning with Large Language Models (ICLR 2025 · similarity 0.90)
- LongCoT: Benchmarking Long-Horizon Chain-of-Thought Reasoning (ICML 2026 · similarity 0.90)
- GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models (ICLR 2025 · similarity 0.89)
- \$OneMillion-Bench: How Far are Language Agents from Human Experts? (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
