Most benchmarks in the x402 / AI-agent space are marketing in disguise. Ours aren't: same queries, same time window, published method, every number either real or explicitly marked missing. If you re-run our numbers and get something different, say so — a benchmark that can't survive reproduction is just an ad.
New agents kept asking "which endpoint do I need?" — so we shipped a free router: POST your task in plain language to /recommend/, get the top-3 matching endpoints with prices and ready curls. No payment, no key. It is now wired into llms.txt, the fleet SKILL.md and the hub, so agents discover it on their own.
tokenprice ($0.001) vs the price-feed incumbents — measured, not marketedtokenprice at $0.001/call vs CoinGecko / CoinMarketCap ($29–129/mo) and CMC's own x402 path at $0.01/call. Live measurements on 4 queries (~1.8s median), cost math at agent scale — plus the honest caveat we could have hidden: ticker lookup is ambiguous across chains (a Solana token can outrank real ETH on the symbol "ETH"). Query by {chain, address} when precision matters.
The method (steal it: fixed query set, same time window, blind scoring, cost logged per query) and our own baseline: real latency and cost from 7 live endpoints. Sub-2.5s across the board; 1,000 news queries = $2.00. Transparent pricing for 10 core endpoints, no "contact sales".
websearch vs Exa vs Tavily — the head-to-head, once we have the keys to run it fairly. We'd rather delay than fake it.X-Free-Remaining header.