Research map / Large language models / Agents, code and math
LocalBench: Benchmarking LLMs on County-Level Local Knowledge and Reasoning
Zihan Gao, Yifei Xu, Jacob Thebault-Spieker
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 web search to GPT-4.1 (ablation) | GPT-4.1 · LocalBench non-numerical QA | GPT Judge accuracy: 47.0 -> 35.6 | Hurt |
| add search grounding to Gemini-2.5-Pro (ablation) | Gemini-2.5-Pro · LocalBench non-numerical QA | GPT Judge accuracy: 52.5 -> 56.8 | Improved |
| add web search to GPT-4.1 (ablation) | GPT-4.1 · LocalBench numerical QA | numerical accuracy: 6.2 -> 15.5 | Improved |
| add search grounding to Gemini-2.5-Pro (ablation) | Gemini-2.5-Pro · LocalBench numerical QA | numerical accuracy: 12.8 -> 12.8 | Mixed |
| add web search to GPT-4.1 (ablation) | GPT-4.1 · LocalBench non-numerical QA | answer rate: 100.0 -> 92.9 | Hurt |
| add search grounding to Gemini-2.5-Pro (ablation) | Gemini-2.5-Pro · LocalBench non-numerical QA | answer rate: 100.0 -> 91.7 | Hurt |
| scale Qwen3 from 32B dense to 235B-A22B MoE | Qwen3 · LocalBench non-numerical QA | GPT Judge accuracy: 27.7 -> 27.3 | Hurt |
| scale Qwen3 from 8B to 14B | Qwen3 · LocalBench non-numerical QA | GPT Judge accuracy: 22.9 -> 27.5 | Improved |
| scale Qwen3 from 14B to 32B | Qwen3 · LocalBench non-numerical QA | GPT Judge accuracy: 27.5 -> 27.7 | No clear effect |
| use newer GPT-4.1 instead of GPT-4o | GPT-4o · LocalBench non-numerical QA | GPT Judge accuracy: 32.8 -> 47.0 | Improved |
| scale Qwen3 from 8B to 235B-A22B on numerical QA | Qwen3 · LocalBench numerical QA | numerical accuracy: 3.1 -> 6.6 | No clear effect |
Closest papers by meaning
- GeoLLM: Extracting Geospatial Knowledge from Large Language Models (ICLR 2024 · similarity 0.88)
- BLEnD: A Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languages (NeurIPS 2024 · similarity 0.86)
- GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing (CVPR 2026 · similarity 0.86)
- INCLUDE: Evaluating Multilingual Language Understanding with Regional Knowledge (ICLR 2025 · similarity 0.85)
- MDK12-Bench: A Multi-Discipline Benchmark for Evaluating Reasoning in Multimodal Large Language Models (AAAI 2026 · similarity 0.85)
- metabench - A Sparse Benchmark of Reasoning and Knowledge in Large Language Models (ICLR 2025 · similarity 0.85)
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.
