On this page · 5 sections ▾
Tokens are the small chunks of text an AI model reads and writes, and they are the unit most AI companies use to bill. Pricing "per million tokens" means a fixed charge for every million chunks sent in (input) or generated back (output). When Anthropic launched Claude Fable 5.2, it set the rate at $8 per million input tokens and $40 per million output tokens, as The Next Web reported.
What is a token?
A language model does not read words the way people do. It splits text into pieces, called tokens, which can be a whole short word, part of a longer word, or punctuation. Common background knowledge puts a token at roughly three-quarters of an English word, though this varies by language and content.
Every request has two sides. The input is everything the model is given: your question, any documents, earlier conversation, instructions. The output is what the model writes back.
Why do input and output cost different amounts?
In this launch, output is five times the price of input: $40 against $8. Reading a prompt can be done largely in parallel, while writing an answer happens one token at a time, with each new token depending on the ones before it. That sequential work ties up expensive hardware for longer, so providers charge more for it.
This matters for planning. A task that feeds in a huge codebase but returns a short summary costs very differently from one that generates thousands of lines of new code.
What changed with Fable 5.2?
According to Digital Trends, the prices fell from $10 and $50, which works out to a 20% cut on both sides. Anthropic also says the new models generate responses up to 30% faster than the 5.1 versions.
There is also an optional fast mode that produces tokens up to 2.5 times quicker. Per The Next Web, it costs $16 and $80, exactly double the standard rates. Speed, in other words, is a separate product you can buy: the same model, served with more computing resources per request.
Why do tokens matter for agent work?
The launch examples involve far more than a single question and answer. Anthropic says a team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours, and an engineer reportedly had the model coordinate more than 1,000 agents over eight days to build an operating system that boots on a laptop.
