Build Show Network
Enter the password to view.
Incorrect password.
Build Show Network · ask.build
An estimate of what it costs to run AI-powered search and brief generation on ask.build, and how that cost behaves as usage grows. Adjust the inputs below to explore scenarios.
Share of searches that run the Sonnet deep dive. 100% reflects today's auto-fire behavior.
Cost per search
$0.029
Monthly — Anthropic
$283
Monthly — Cloudflare
$5
Monthly — total
$288
At your current cache hit rate, comparing today's always-on deep dive against an on-demand model where Sonnet fires on roughly one in five searches.
| Searches / month | Deep dive always on | Deep dive on-demand |
|---|
The Sonnet deep-dive analysis is roughly 85% of the per-search cost and currently runs automatically on every search — including searches where no one ever opens the brief. Firing it only when a user asks to go deeper, or opens a Field Brief, is the single biggest cost reduction available, with no change to the instant first answer.
Identical questions are served from cache at effectively zero AI cost. The higher the share of searches answered from cache, the lower the bill. Light normalization of incoming queries raises this share at no meaningful cost.
The fixed instructions sent on every request can be cached at the model provider, cutting that portion of the input cost by up to 90%. It compounds with the levers above and is not reflected in the figures here, so real-world cost is likely to come in lower than shown.
Each search runs a fast first pass (Claude Haiku) for the immediate answer, and a deeper second pass (Claude Sonnet) for the richer analysis that backs the brief. The brief itself reuses that already-generated analysis, so it carries no separate AI charge. Cached searches incur no model cost.
Costs fall into two buckets. The Anthropic API (the two model passes) is the overwhelming majority of marginal cost. Cloudflare infrastructure — vector search, edge compute, and the database — is a small, near-fixed amount that stays modest even at a million searches a month.
Model prices are Anthropic's published per-million-token rates as of June 2026 (Haiku 4.5 at $1 / $5, Sonnet 4.6 at $3 / $15) and Cloudflare's published Workers, Vectorize, and D1 rates. Token-per-search figures are working estimates and are intended to be replaced with measured averages from production usage; the calculator's assumptions panel is provided for exactly that. Figures are planning estimates, not a quote.