Headless ThronesA LIVING HISTORY OF AI
Loading recorded history
The wireIndependent of map date

Loading the latest source headlines…

All news ↗
A LIVING HISTORY OF AI

This week in the AI world.

AI companies are kingdoms. Their share of recorded token usage shapes the land.

Loading recorded source history…OpenRouter text-token activity shapes territory. Buildings and street life are illustrations.

The living worldLoading recorded history…
THE OPEN OCEANA world in the making
CONTINENT1.0×Scroll or pinch to explore
Settled borders follow token activity
Charting the continent…
Loading…Loading…
Loading…RECORDED DATES
A WORLD THAT REMEMBERS

What changed?

Explore the dated events behind the places.

Loading the world’s history…

Recorded Source-backed facts Illustrated World effects Reconstructed Earlier development

THE HEADLESS THRONES JOURNAL

News from the AI world

Source headlines link to their publishers. Our original, agent-assisted stories are labeled Headless Thrones.

View all stories
✎

The newsroom is opening.

External headlines and published Headless Thrones stories will appear here, with attribution and dates.

Visit the journal →

All kingdoms
THE MODEL ALMANAC

Every building has a story.

Loading real model identities and their recorded activity through the selected date. Missing daily top-50 rows are unknown, not zero.

Model & familyKingdom7-day observed tokensDays observedMeasurement
A FIELD GUIDE

How this world works.

A visual history of AI companies. Recorded token activity shapes their territory; illustrated cities make that history a place to explore.

How to read this world
◇ Land & borders
Recorded OpenRouter text-token share
▥ Buildings & transport
Illustrated places and street life
ϟ Outages & repair
Official incidents, illustrated effects
⌁ Unclaimed frontier
Designed room for future kingdoms

Recorded activity. Illustrated life. Borders use a platform sample, not worldwide market share. Events link to sources. Building designs, people, vehicles and visual repairs are illustrations; earlier development is reconstructed. Population, sentiment, allowances and finances are not measured here.

…publisher identities
…real model identities
…unclaimed frontier
01

Recorded activity becomes land

Each kingdom represents a model publisher. Seven consecutive complete UTC days of reported OpenRouter text tokens determine its share of settled land. Undisclosed activity stays in neutral territory. A missing model row means unknown usage, not zero. The green frontier is reserved space and does not represent token demand.

02

Buildings make history visible

Zoom into OpenAI to inspect model facilities and their permanent tile records. The architecture, city size, pedestrians and vehicles are artwork. A skyscraper is not evidence of company funding, revenue, model quality or a measured technology ranking.

03

Sources anchor the events

Official OpenAI incidents connect to a local service facility. Damage and repair visualize those events; they are not measurements of physical destruction. Recorded, illustrated and reconstructed entries are labeled, and the record links to its evidence.

04

Explore a dated record

Move through the timeline, choose a kingdom or tap a place. When token data ends before the selected event date, the map holds the latest complete token window and labels its dates. The news feed has its own publication dates and stays independent of the map date.

A note from the cartographer

The continent and its street life are illustrations. Only recorded source activity determines token shares; no invented usage, sentiment, finances or population estimates fill gaps. More companies and model identities can appear as the source archive grows.
THE DATA BEHIND THE WORLD

A platform sample, clearly labeled.

Colored land Company kingdomsNatural land Unclaimed frontierNeutral land Unattributed activityTap a tile for its history · Pinch or scroll to zoom
… unclaimed Loading land allocation…

What moves the borders

The map uses the OpenRouter daily token dataset, filtered to its text-model surface. Seven consecutive UTC days are added before calculating shares. The source strip shows the selected token window and the latest available archive date. If newer events are available before another complete token window, territory holds its last recorded values; no usage is estimated to fill the gap.

What “other” means

Each source day identifies its top 50 models and combines the rest into “other.” We keep that residual, undisclosed creators and mixed routing in neutral unattributed territory. Named publishers may have additional usage hidden there. Missing rows are unknown, not zero. The reserved green frontier is a separate design feature for future kingdoms.

What the map does not measure

It is not the whole AI market, a count of users or US-only activity. Provider tokenizers differ. Buildings are illustrations; real sentiment, allowances and finances are not connected. Official OpenAI incidents are connected; their effects on buildings are illustrations. Small shares may be below a whole map tile.

Identities and history

Models belong to their publisher, not the company hosting inference. Explicit aliases combine Meta namespaces, Qwen/Alibaba, xAI, Mistral, Moonshot and Z.ai. Fine-tune publishers keep their own identities. Historical gaps remain gaps. The source’s text filter can omit retired aliases; this is the published text surface, not a complete reconstruction of every historical model.

Independent gateway views

These sources measure different platform samples and periods. They publish percentages, not absolute token counts, and do not change the OpenRouter map.

Select a source to inspect its recorded token shares.

Attribution & reuse

Source: OpenRouter (openrouter.ai/rankings). CC BY 4.0. The download contains the exact source timestamp, counts, model identifiers and method.

OpenRouter rankings ↗ · CC BY 4.0 ↗

Token history and news publishing are separate systems. A new headline does not by itself change borders, model usage or city conditions.