Climate-ModernBERT Achieves 76.3 Average F1 on Climate NLP Benchmarks
September 9, 2026
Continued pretraining of ModernBERT-Base on mixed academic, web, and synthetic climate corpora improves average F1 by 2.8 points over the vanilla baseline. The research identifies academic corpora as the most effective driver for domain adaptation in climate-specific NLP tasks.
HOW THIS AFFECTS YOU
●
builderYou can use these specialized weights for more accurate climate-related text classification or extraction.
●
researcherYou can leverage specific corpus mixtures to improve domain-adaptive encoder performance.