An Arena.ai analysis of more than 26,000 sessions involving 30 AI agents from 13 labs found that top-performing agents consume about 93 cached or re-read input tokens for every token they generate. Context re-reads account for 86% to 96% of total token usage, while actual outputs represent only 1% to 2%. Tasks run without caching cost 3.6 times more than the same tasks with caching; one example involving Claude Fable 5.1 rose from $7.44 with caching to $27 without it. Businesses should therefore evaluate the full cost of tasks with caching, while developers should keep context stable and reusable to capture discounts.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
