Sources & credits
Where the numbers come from
Every stat on every card is a published measurement. We don’t make numbers up or estimate them, and when a lab doesn’t publish a figure, the card says so. The snapshot was taken on 2026-10-11.
Benchmarks, speed and price
Intelligence, hallucination rate, context window, output speed, response time, prices and modalities come from Artificial Analysis’s independent benchmark measurements of each model. Every card uses one configuration of its model (shown on the card, for example “Max” reasoning effort), so all of a card’s numbers come from the same setup. That’s why flagship reasoning models have long response times. Prices are blended 3:1 input to output; a few use the public list price from OpenRouter.
Knowledge cutoffs
From each lab’s own model docs or model cards, linked below. One date comes from a secondary source and is marked as such. Labs that don’t publish a cutoff get N/D.
Art, sound & logos
Card illustrations and shop photos were generated with StickyPrompts image generation. Music and sound effects were generated with ElevenLabs. Lab logos are from LobeHub Icons (MIT). Visitor locations in our own statistics use IP Geolocation by DB-IP (CC BY 4.0). All model names and logos are trademarks of their owners. Top Tokens is a fan-made game and isn’t affiliated with or endorsed by any AI lab.
Knowledge cutoff sources
* secondary source