Research map / Large language models / Agents, code and math
TaxReasoning: Benchmarking Knowledge-Intensive Mathematical Reasoning with Evolving Tax Laws
Nan Hu, Yike Wu, Jiaye Li, HuiKang Hu, Guilin Qi, Songlin Zhai, Yongrui Chen, Tianxing Wu, Tongtong Wu, Jiaoyan Chen, Jeff Z. Pan
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 |
|---|---|---|---|
| add standard retrieval of tax documents to CoT | Gemini 2.5 Pro · TaxReasoning | accuracy (%): 31.37 -> 57.22 | Improved |
| add oracle gold documents to CoT | GPT-4o · TaxReasoning | accuracy (%): 12.75 -> 14.76 | No clear effect |
| add standard retrieval to CoT | Qwen2.5-72B · TaxReasoning | accuracy (%): 17.65 -> 21.51 | Improved |
| use PoT instead of CoT prompting | DeepSeek-R1 · TaxReasoning | accuracy (%): 42.16 -> 38.00 | Hurt |
| use PoT instead of CoT prompting | OpenAI o3 · TaxReasoning | accuracy (%): 19.61 -> 21.57 | Improved |
| summarize retrieved documents instead of standard retrieval | Gemini 2.5 Pro · TaxReasoning | accuracy (%): 57.22 -> 56.33 | Hurt |
| use InlineSearch instead of Interact multi-turn retrieval | DeepSeek-R1 · TaxReasoning | accuracy (%): 53.75 -> 59.48 | Improved |
| use multi-turn retrieval (Interact) instead of single-turn Standard | GPT-4o · TaxReasoning | accuracy (%): 11.67 -> 13.01 | Mixed |
| use Interact instead of Standard RAG | Qwen2.5-72B · TaxReasoning | accuracy (%): 21.51 -> 17.14 | Hurt |
| retrieve from a structured tax knowledge graph instead of raw documents | DeepSeek-R1 and Gemini 2.5 Pro · TaxReasoning | accuracy (%) | Improved |
| give the model open-book human access vs closed-book (humans) | human CPAs · TaxReasoning (50-question sample) | accuracy (%): 61.00 -> 82.00 | Improved |
Closest papers by meaning
- GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models (ICLR 2025 · similarity 0.87)
- Putnam-AXIOM: A Functional & Static Benchmark for Measuring Higher Level Mathematical Reasoning in LLMs (ICML 2025 · similarity 0.87)
- A Careful Examination of Large Language Model Performance on Grade School Arithmetic (NeurIPS 2024 · similarity 0.87)
- RealMath: A Continuous Benchmark for Evaluating Language Models on Research-Level Mathematics (NeurIPS 2025 · similarity 0.87)
- SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models (ICML 2024 · similarity 0.87)
- UGMathBench: A Diverse and Dynamic Benchmark for Undergraduate-Level Mathematical Reasoning with Large Language Models (ICLR 2025 · similarity 0.86)
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.
