Cost¶
What the whole live suite cost, and how to read the per-request numbers.
The live suite on 2026-09-29 made 210 requests over 17 tools and 184 labelled cases: 135,192 input tokens, $0.005677 at $0.042 per million input tokens, output free. That is about 0.6 cents for every number on this site.
Per request¶
Mean latency across the suite was 175 ms per request, measured on the client from a laptop, network included. Latency follows prompt length, not question count: many questions over one state cost about what one question costs, which is why every tool asks everything it needs in one request.
Per tool¶
The function_call rows are the expensive ones on the table: 30 questions per command, so a 16-command run used more tokens than the other four tools together. That is the trade the cookbook makes on purpose, one request instead of a round trip per argument. The composite_score rows are the cheap ones, one request per item with three questions in it.
Watching it¶
jev_usage reads the tally: calls, questions, input tokens, mean latency, cost so far. Every result also carries input_tokens and latency_ms for its own request. The price constant is PRICE_PER_MILLION_INPUT_TOKENS in strands_jev/questions.py; when TypeSafe changes the price, that one line changes and every cost on this site with it.