Saturday · 9 May · 2026“Every chart is an opinion in a coat of evidence.”Interactive Edition
Interactive · Companion to The Index
Seven Anatomies of an Interactive Chart
A small grapher in the spirit of Our World in Data — built from
the same primitives every modern data editorial uses, hand-rolled
in vanilla SVG. Drag the sliders, hover the lines, toggle the
scales. Each anatomy is a different argument the chart is allowed
to make.
Anatomy 01 · Line chart with hover & log toggle
Training tokens consumed by frontier models, 2018–2026
Cumulative training tokens, in trillions. Hover for values, click a series in the legend to mute it, switch to log to see the early years.
Y-axis
The Atlas · 2026Indicative · reconstructed from public model cards · CC-BY 4.0
Anatomy 02 · Discrete bar with year slider & sort
Vocabulary size by model
Tokens in each model's vocabulary, in thousands. Drag the year to see how vocabularies grew. Press play to animate.
2026
Sort
The Atlas · 2026Source: Tokenizer configs · Indicative
Anatomy 03 · Stacked area with relative toggle
Pre-training corpus mix, 2018–2026
Composition of the training data fed to a typical frontier model, by source. Switch to relative to read each layer as a share of the total.
Mode
The Atlas · 2026Reconstructed corpus mixes · Indicative
Anatomy 04 · Animated scatter with bubble size
Capability vs cost, by year
Each bubble is a model. X is benchmark score, Y is dollars per million tokens, bubble size is parameter count. Press play to watch the frontier move.
2026
Y-axis
The Atlas · 2026Indicative composite of public benchmarks & price sheets
Anatomy 05 · Slope chart between two years
How much each language gained on the open web, 2020 → 2026
Share of public web text, by language, at two points. Lines that rise gained ground; lines that fall lost it.
The Atlas · 2026Indicative web-text share · all values approximate
Anatomy 06 · Stacked bar over time
Where the compute went, 2019–2026
Estimated training compute attributed to major labs each year, in petaflop-days, stacked.
The Atlas · 2026Indicative · public reports & estimates
Anatomy 07 · Choropleth map (grid form)
Web-text share, by region (2026)
Each cell is a region. Darker means a larger share of the public web is written there. Hover any cell.
0%≥ 50%
The Atlas · 2026Indicative · grid stand-in for true geographic projection