BEND 2.0 Benchmark Evaluates DNA Sequence Design Capabilities
November 2, 2026
BEND 2.0 measures genomic foundation models on DNA sequence prediction and design across eight tasks. Findings show that linear readouts on frozen representations can match native model outputs and that functional post-training significantly improves sequence design efficacy.
HOW THIS AFFECTS YOU
●
researcherUse this benchmark to evaluate if models capture functional patterns rather than just sequence statistics.
●
healthThis provides a more rigorous standard for validating AI tools used in genomic engineering.