0.75
#1 of 4
AUPRC on variant effect prediction for GPN-Star (200M parameters, 2025), against 0.53 for Evo 2 (40B)

What the chart shows

Genomic models try to predict which DNA variants change how genes work. In results reported by Ye et al. (2025), GPN-Star, with 200 million parameters, scored 0.75 on the area under the precision–recall curve (AUPRC), against 0.53 for Evo 2, which has 40 billion.

The older Borzoi (2023) and Enformer (2021) scored 0.40 and 0.38. The AI Index cites the result as evidence that scale alone is not yet enough for virtual cell and genomic models, and that training method and data matter.

Key numbers

MeasureValue
GPN-Star (200M, 2025)0.750
Evo 2 (40B, 2025)0.525
Borzoi (190M, 2023)0.400
Enformer (250M, 2021)0.375
Evo 2 parameters ÷ GPN-Star parameters200×

How to read it

  • AUPRC runs from 0 to 1; higher is better. The AI Index does not say which variant set the scores come from.
  • Parameter counts in labels are as the AI Index gives them; the data file lists Enformer at 246 million and Borzoi at 186 million.
Source: Ye et al., 2025, via the AI Index Report 2026 (Stanford HAI), figure 6.1.12: Virtual cell model performance. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart