The state of biological AI at a glance — which models drive the most academic impact and developer adoption, and how open the field actually is. Across 926 models, with 923 evaluated against the Model Openness Framework.
Every published model placed by what it does — models with similar descriptions sit close together, so the clusters are the field's real subfields rather than our category labels.
926 models positioned by what their descriptions say — similar models sit close together. Scroll to zoom, drag to pan, click a point to open it.
The map above places all 926 published models by what their descriptions say, so models solving similar problems form clusters. It is a visual view; the most-cited models in each area are listed below.
The biggest and most-cited models — colored by openness so you can see whether scale and openness go together.
How model scale relates to academic citations (log-log). Showing 167 models with both metrics.
Academic impact by citation count
| # | Model | Citations |
|---|---|---|
| 1 | 37.6K | |
| 2 | 12.4K | |
| 3 | 5.1K | |
| 4 | 3.6K | |
| 5 | 3.2K | |
| 6 | 3K | |
| 7 | 2.7K | |
| 8 | 2.4K | |
| 9 | 1.9K | |
| 10 | 1.9K |
Developer adoption by GitHub stars
| # | Model | Stars |
|---|---|---|
| 1 | 15.1K | |
| 2 | 14.8K | |
| 3 | 14.8K | |
| 4 | 8.3K | |
| 5 | 4.7K | |
| 6 | 4.7K | |
| 7 | 4.5K | |
| 8 | 4.4K | |
| 9 | 4.2K | |
| 10 | 4.2K |
MOF's pass/fail tiers leave most models unclassified, so the bio.rodeo openness score rates them on a 0–100 gradient — split into usability (can I run it?) and reproducibility (can I retrain it?). These views show that gradient, the trend over time, where models fall short, the licenses in play, and how openness relates to impact.
Each dot is a model; color is its bio.rodeo openness band. Below the diagonal: easier to run than to retrain (open weights, closed recipe); above it: more reproducible than usable. Showing 923 evaluated models.
Top models by bio.rodeo openness score (usability + reproducibility, license-aware).
| # | Model | Use | Repro | Overall |
|---|---|---|---|---|
| 1 | Beijing Academy of Artificial Intelligence | 100 | 100 | 100 |
| 2 | Chan Zuckerberg Biohub / Mehta Lab | 100 | 95 | 98 |
| 3 | IOCB Prague / MIT | 100 | 92 | 98 |
| 4 | MRC Laboratory of Molecular Biology / University of Cambridge | 100 | 92 | 98 |
| 5 | MIT | 100 | 95 | 97 |
| 6 | Microsoft Research | 100 | 92 | 96 |
| 7 | Broad Institute / Dana-Farber Cancer Institute | 100 | 93 | 96 |
| 8 | Chan Zuckerberg Initiative | 100 | 92 | 96 |
| 9 | Chan Zuckerberg Initiative | 100 | 92 | 96 |
| 10 | Aikium | 99 | 92 | 96 |
Evaluated models by release year and overall openness tier (open / partial / closed)
Share of models that release each MOF component openly, grouped by pillar — where the field falls short
Median citations for models that release a component openly vs. not, sorted by the size of the gap
Most common licenses across released components, colored by whether they are open
Openness and impact broken down by biological domain and by organization, plus the field's growth in scale over time.
Models per biological domain, split into open (Class III+) and not open
By total citation count
How model scale has grown over time, colored by bio.rodeo openness score. Bubble size reflects citation count where available. Showing 227 models.