Say what you are building in plain language. It reads your text for domains, agentic tasks and constraints, recommends an open-weight model stack from the verified benchmark maps, and estimates cost per task — the number that actually shows up on the invoice.
Keyword-based, not a live model. Remove a chip you did not mean, or add one it missed — the recommendation updates immediately.
Every number here is editable and saved in your browser.
Defaults are approximate blended serverless list prices (mid-2026). Paste your provider’s real input and output numbers and every cost updates. Reset to defaults.