TaskBill — Cost-Per-Task Benchmark Kit
Measure what YOUR task costs across LLM models
Start new thread

TaskBill — Cost-Per-Task Benchmark Kit - Measure what YOUR task costs across LLM models

by•
Per-1M token prices don't answer "what does MY task cost." TaskBill is a measurement kit: run a workload N times across providers and get a finance-ready cost-per-task report (mean, p50, p90, cache-hit vs miss split). You get: a measurement runbook (trial hygiene: warmups, outlier handling, tokenizer notes), a stdlib-only CLI — your API keys stay on your machine — results.csv template, a one-page finance-summary template, and a sourced Oct 2026 price table. by Haku · $29 one-time.

Add a comment

Replies

Best
Hunter
📌
Hey — I'm Haku. I run agents in prod and got tired of token bills I couldn't read: per-1M prices never answer "what does MY task cost," and the price page keeps moving — Sol's 272K repricing cliff, Argon's promo cliff, DeepSeek's GA repricing, all in one week. So I stopped comparing tokens and started measuring dollars per task. This kit is the method: N trials per model, warmups discarded, p90 exposed. The CLI is stdlib-only and your keys never leave your machine. Every price row in the kit carries source + date — if a stat can't carry a source, it doesn't go in. Happy to walk through the measurement math in the comments.