521M
median parameters of the seven protein structure prediction models released in 2025; AlphaFold 3 (2024) has 370M
Median size by release year
20212025
What the chart shows
After AlphaFold 3 (370 million parameters, May 2024), structure prediction models stayed at a similar scale. The seven 2025 releases in the data range from 109 million (Protenix-Tiny) to 3 billion (SimpleFold), with a median of 521 million (Boltz-2).
ESM3, a 98-billion-parameter protein language model that also predicts structure, is the outlier. AlphaFold 2 and RoseTTAFold, both from July 2021, had 93 million and 14 million parameters. The AI Index argues that data, more than model size, now limits progress (related chart).
Key numbers
| Measure | Value |
|---|---|
| AlphaFold 2 (Jul 2021) | 93M |
| AlphaFold 3 (May 2024) | 370M |
| ESM3 (Jul 2024) | 98B |
| Boltz-2 (Jun 2025) | 521M |
| SimpleFold (Jul 2025) | 3B |
| Protenix-Tiny (Jul 2025) | 109M |
How to read it
- Selected models compiled by RAISE Health; not a complete list.
- The vertical axis is logarithmic. Hover over a dot to see the model.
- The median is our calculation from the seven models dated 2025 in the data.
Source: RAISE Health, 2026, via the AI Index Report 2026 (Stanford HAI), figure 6.1.8: Size of protein structure prediction models, 2021–25. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion