Research map / Large language models / Agents, code and math
RefTool: Reference-Guided Tool Creation for Knowledge-Intensive Reasoning
Xiao Liu, Da Yin, Zirui Wu, Yansong Feng
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 RefTool tools to PoT | PoT · QRData-causal | accuracy: 37.1 -> 41.6 | Improved |
| add RefTool tools to ReAct | ReAct · QRData-causal | accuracy: 43.8 -> 46.3 | Improved |
| add RefTool tools to PoT | PoT · TheoremQA-physics | accuracy: 52.2 -> 54.8 | Improved |
| add RefTool tools to PoT | PoT · SciBench-chemistry | accuracy: 55.0 -> 57.7 | Improved |
| add plain RAG over the same textbooks to PoT | PoT · QRData-causal | accuracy: 37.1 -> 36.4 | No clear effect |
| add plain RAG over the same textbooks to PoT | PoT · TheoremQA-physics | accuracy: 52.2 -> 51.1 | Hurt |
| add plain RAG over the same textbooks to PoT | PoT · SciBench-chemistry | accuracy: 55.0 -> 55.6 | No clear effect |
| add code-form tools in place of raw text segments (hierarchy kept) (ablation) | PoT + RefTool · QRData-causal | accuracy: 39.0 -> 41.6 | Improved |
| add code-form tools in place of raw text segments (hierarchy kept) (ablation) | PoT + RefTool · TheoremQA-physics | accuracy: 52.2 -> 54.8 | Improved |
| add hierarchical selection in place of similarity-based retrieval (ablation) | PoT + RefTool · QRData-causal | accuracy: 38.0 -> 41.6 | Improved |
| add hierarchical selection in place of similarity-based retrieval (ablation) | PoT + RefTool · TheoremQA-physics | accuracy: 51.8 -> 54.8 | Improved |
| add RefTool to Zhuang-Chinese translation prompting | Rule-by-Rule Retrieval w. Code · Zhuang-Chinese translation (ZhuangRules) | BLEU: 43.9 -> 54.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TroVE — TroVE: Inducing Verifiable and Efficient Toolboxes for Solving Programmatic Tasks (ICML 2024) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Automated Creation of Reusable and Diverse Toolsets for Enhancing LLM Reasoning (AAAI 2025 · similarity 0.91)
- ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs (ICLR 2024 · similarity 0.88)
- AutoTool: Dynamic Tool Selection and Integration for Agentic Reasoning (ICML 2026 · similarity 0.88)
- ReTool: Reinforcement Learning for Strategic Tool Use in LLMs (ICLR 2026 · similarity 0.88)
- Large Language Models as Tool Makers (ICLR 2024 · similarity 0.88)
- Joint Optimization of Tool Creation and Use for Large Language Model Agents (NeurIPS 2026 · similarity 0.87)
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.
