A live calculator for AI teams debating model performance and ROI. Set token prices, run volume, error rate and retry policy to see daily and monthly spend, a run timeline with retries, and bottleneck flags instantly. Numbers here are illustrative estimates, not financial statements about any company, and nothing on this page is investment advice.
Simulated 40-run sample at your error rate. Bar height is tokens per attempt; stacked segments above a bar are retry attempts triggered by failures.
0 of 8 checks confirmed.
The default attempt uses four thousand prompt tokens at zero point zero zero three dollars per thousand and twelve hundred completion tokens at zero point zero one five. Those contribute zero point zero one two and zero point zero one eight dollars, totaling three cents. At eight thousand pixels per dollar prompt cost measures ninety six, completion one hundred forty four and total two hundred forty. The assumptions are fixed prices and average token counts, not live provider billing. Every retry is charged the same amount rather than modeling growing conversation context or a cheaper recovery model. With independent eight percent failure and two maximum retries, expected attempts are one plus zero point zero eight plus zero point zero zero six four, or one point zero eight six four. At two hundred pixels per expected attempt the first contribution measures two hundred, first retry sixteen and second retry one point two eight. Final failure probability is eight percent cubed, or zero point zero five one two percent. The display named cost per successful run uses cost per initial run including failures; it does not divide by the success probability. Independence and equal retry cost are assumptions, not observed reliability. Multiplying three cents by one point zero eight six four gives zero point zero three two five nine two per initial run. At five hundred runs daily spend is sixteen point two nine six dollars versus fifteen with no retries, a difference of one point two nine six. At fifteen pixels per dollar these bars measure two hundred forty four point four four, two hundred twenty five and nineteen point four four. Thirty days gives four hundred eighty eight point eight eight dollars. The forty-run visual timeline uses a fixed pseudo-random seed, while checklist counts are self-reported signals. Native presets, inline SVG redraw, checklist and reset operate locally and never prove actual workers, measured failures or a production budget.