AI inference cost simulator

Explore worked analyses of real AI products, or model your own: build it by hand or paste it in from your AI coding assistant. Set any input as fixed or simulated, then read the probability that your inference cost stays below a level you set.

derived from a public open-source research agent's code and prompts (config defaults, template sizes); volumes are estimates, not logs

SummaryYour product’s inference calls cost between $0.178 and $0.277 per report 80% of the time (median $0.225).

Set your offer pricing to read cost as a share of revenue against the 20% margin benchmark. Tap on 🔮 to add an input to a what-if scenario to explore ways to optimize your cost or margin.

Probability density|

Percentage of 2,000 draws falling below/above unit cost

$0$0.1$0.2$0.3base median $0.224745%55%$0.22

Probability % margin of error is ±1.1 pts due to sampling error.

Cost breakdown
task$ per report% of costKtok
research report (basic flow)$0.19100%13.7
blended $14.197 per Mtok
Sensitivity analysis

Impact of calibrating a variable to its typical value

$0.20$0.25$0.30$0.29nowa · report writer (full scraped context) · output tokens per call — $0.2930 → $0.2360 · alone: $0.2930 → $0.2360ab · report writer (full scraped context) · input tokens per call — $0.2360 → $0.1945 · alone: $0.2930 → $0.2822b$0.19all

each step: the same 2,000 draws re-run with that variable and every one before it held at its typical; to pin a variable down, measure it, test it, or cap it

Your offer pricing
Tasks & calls
taskfrequencycallssimulated
research report (basic flow)1 run per report3 calls per run · frontier ×1 · workhorse ×2
Simulated variables
report writer (full scraped context) · input tokens per call
400010k16k
report writer (full scraped context) · output tokens per call
130017004000

Variables are sampled independently (no correlation between them) with a fixed seed; the base and what-if curves see the same samples, so their differences are the decision's effect alone.