MarinDNA 1B Model Matches Evo 2 40B Performance with 1980x Less Compute

anshulkundaje · x · 2026-08-19

MarinDNA shows a 1B parameter standard Transformer rivals Evo 2 40B on variant effect prediction (VEP) using curated data and scaling laws. It uses 1,980x fewer training FLOPs and scores variants 2,330x faster. The study notes zero-shot VEP may regress with scale while linear-probe VEP improves.

Original post →

More from Research

Research channel →