Research map / Large language models / Agents, code and math
MathNet: A Global Multimodal Benchmark for Mathematical Reasoning and Retrieval
Shaden Alshammari, Kevin Wen, Abrar Zainal, Mark Hamilton, Navid Safaei, Sultan Albarakati, William T. Freeman, Antonio Torralba
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 expert-paired related problem and solution (Expert-RAG) vs zero-shot (ablation) | DeepSeek-V3.2-Speciale · MathNet-RAG (human grading) | accuracy: 84.8 -> 97.3 | Improved |
| add expert-paired related problem and solution (Expert-RAG) vs zero-shot (ablation) | GPT-5 · MathNet-RAG (human grading) | accuracy: 76.8 -> 86.6 | Improved |
| add embedding-retrieved problem and solution (Embed-RAG) vs zero-shot (ablation) | GPT-5 · MathNet-RAG (human grading) | accuracy: 76.8 -> 75.2 | No clear effect |
| add embedding-retrieved problem and solution (Embed-RAG) vs zero-shot (ablation) | Grok-4.1-Fast · MathNet-RAG (LLM grading) | accuracy: 73.1 -> 67.7 | Hurt |
| add embedding-retrieved problem and solution (Embed-RAG) vs zero-shot (ablation) | Gemini-3-Pro · MathNet-RAG (LLM grading) | accuracy: 73.2 -> 70.5 | Hurt |
| use expert-paired instead of embedding-retrieved example (ablation) | Gemini-3-Pro · MathNet-RAG (human grading) | accuracy: 92.9 -> 87.5 | Hurt |
| add expert-paired example (Expert-RAG) vs zero-shot (ablation) | Claude-4.5-Opus · MathNet-RAG (human grading) | accuracy: 46.8 -> 52.4 | Mixed |
| add expert-paired example (Expert-RAG) vs zero-shot (ablation) | Phi-4-Reasoning Plus · MathNet-RAG (human grading) | accuracy: 15.1 -> 16.7 | No clear effect |
| use harder retrieval tier (Hard vs Easy positives) (ablation) | gemini-embedding-001 · MathNet-Retrieve | Recall@5: 90.68 -> 9.31 | Hurt |
| use newer/larger embedding model (gemini-embedding-001 vs text-embedding-ada-002) | MathNet-Retrieve · MathNet-Retrieve (All avg) | Recall@5: 40.72 -> 59.02 | Improved |
| reasoning-capable model vs non-reasoning (gemini-3.1-pro vs gpt-4o) | MathNet-Solve-Test · MathNet-Solve-Test | accuracy: 6.8 -> 78.4 | Improved |
Closest papers by meaning
- MathSE: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning (AAAI 2026 · similarity 0.92)
- Advancing Mathematical Reasoning in Language Models: The Impact of Problem-Solving Data, Data Synthesis Methods, and Training Stages (ICLR 2025 · similarity 0.91)
- MV-MATH: Evaluating Multimodal Math Reasoning in Multi-Visual Contexts (CVPR 2025 · similarity 0.91)
- UGMathBench: A Diverse and Dynamic Benchmark for Undergraduate-Level Mathematical Reasoning with Large Language Models (ICLR 2025 · similarity 0.90)
- Omni-MATH: A Universal Olympiad Level Mathematic Benchmark for Large Language Models (ICLR 2025 · similarity 0.90)
- RealMath: A Continuous Benchmark for Evaluating Language Models on Research-Level Mathematics (NeurIPS 2025 · similarity 0.90)
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.
