Guides
These guides explain the mechanics behind the numbers on this site: what a token actually is, how prompt caching changes what you pay per request, and exactly how our counters produce their numbers for each provider.
Token counts are not interchangeable across providers. OpenAI's o200k_base tokenizer is open source, so exact GPT counting can run in a browser. Anthropic publishes no tokenizer for current Claude models, and Claude's tokenizer differs by model generation. A count that is exact for one model can be wrong for another. The guides below cover why, and what to do about it.
What is a token?
Tokens and BPE in plain language, the 0.75-words rule of thumb and when it breaks, and why counts differ per provider and per Claude generation.
Prompt caching
Read and write pricing, break-even math for the 5-minute and 1-hour TTLs, and the silent mistakes that invalidate a cache without any error.
How we count
The per-provider counting methods behind our calculators, the 100-prompt verification corpus, and where exact counting is not possible.
Where the numbers live
The token calculator gives you counts for your own text. The pricing pages hold current per-token rates by provider and model, including cached-input and batch rates where a provider offers them.
Guide content is reviewed against provider documentation. Figures on these pages were verified 2026-08-10.