Geo-Expert: Parameter-Efficient LLMs for Geological Reasoning

Researchers introduce Geo-Expert, a family of parameter-efficient large language models fine-tuned for expert-level geological reasoning. The models are built by fine-tuning three base architectures—Qwen3-8B, Qwen3-32B, and Gemma-3-27B—using Low-Rank Adaptation (LoRA) on a custom-curated instruction dataset. A novel benchmark, Geo-Eval, was developed for evaluation. Results show that a domain-aligned 8B model outperforms open-weight 70B generalists and proprietary GPT-4o on specialized geological tasks, while a 32B variant approaches frontier reasoning models. The work addresses the gap where general-purpose LLMs hallucinate on subsurface and deep-time geological reasoning, and current AI in Earth sciences focuses on surface remote sensing and GIS.