AI Fervour — AI chatbots and assistant models compared | aifervour.com
How much text fits in one window? Token limits for the major models, converted into words and pages.
Three numbers define the current field. GPT-4o accepts 128,000 tokens of context, Claude Sonnet 200,000, and Gemini 1.5 Pro 1,000,000. Those are the ceilings on how much text each model can hold at once — your paste plus its reply.
Tokens convert cleanly into words at the published average of 0.75 words per token. The three windows therefore hold about 96,000, 150,000 and 750,000 English words respectively.
On narrow screens, swipe or scroll the plate sideways.
In page terms, a 500-page book is roughly 200,000 tokens. It fills one Claude Sonnet window exactly, needs 1.6 GPT-4o windows, and takes up about a fifth of the Gemini 1.5 Pro window.
The ceiling moved fast. GPT-3 in 2020 capped context at 2,048 tokens; GPT-4 shipped in 2023 with 8,000- and 32,000-token windows; Gemini reached 1,000,000 in 2024.
To use the comparison, multiply the word count of your largest real job by 1.3 and check the result against the windows above. The homepage table places these sizes next to prices, and the token-pricing page shows what one filled window costs to run.
Further reading