A unified geo-benchmark of 421k questions across knowledge, reasoning, and application tasks, showing that thinking mode can help small models close the gap with much larger ones.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
GeoBenchLLM: A Comprehensive Benchmark for Evaluating LLMs on Geo-Related Tasks
A unified geo-benchmark of 421k questions across knowledge, reasoning, and application tasks, showing that thinking mode can help small models close the gap with much larger ones.