Foxxe Labs Research · Interactive
The Expanding Universe of Large Language Models
Every landmark model since the 2017 Transformer, plotted on the city of the lab that built it — sized by parameters, raised by capability, coloured by maker. Press play to grow the map from Google Brain's original paper to September 2026. Drag to spin the globe, click any marker for its detail card, and scrub the timeline to any month — or drag the slider straight to the date you care about.
- Models
- 0
- Labs
- 0
- Countries
- 0
- Largest
- —
Latest releases
Makers
Drag to spin. Click a marker for its card. Filled markers are closed weights, rings are open.
Reading the globe
- Position — the city of the lab that released the model, jittered slightly so co-located releases stay separable.
- Marker size — parameter count, on a square-root scale. Sizes for closed models are best public estimates.
- Spike height — a capability score normalised to 0–100 across the whole set. It is a positioning aid, not a benchmark result.
- Colour — the maker. Filled markers are closed weights, rings are open weights.
- Sound — off until you turn it on. Each model blips as it appears, pitched by capability and panned to where it sits on the globe.
Sizes and scores are approximate and, for closed models, estimated. For live benchmark tracking see the AI Model Performance Timeline or llm-stats.com.