AI task cost calculator

Estimate what a workflow costs to run with AI. Task-type presets give a starting tokens per task for agentic workloads; review is costed as review hours per task × review rate. In production, API inference is often only 15–25% of total AI cost; maintenance, build, and review matter too.

TCA: Total Cost of AI: Monthly AI cost: inference, review, risk premium, maintenance, and amortized build.
Tokens per task: Default token volume for the selected task type (agentic). Replace with pilot-measured tokens when you have real data.
Cost per task: TCA divided by the volume handled on the AI path — a pure unit cost, not a labor-replacement claim.

Adjust model, review hours and rate, risk premium, maintenance, build cost, and automation rate below. All figures are per month unless noted.

Task cost estimate (STH)

Estimate TCA (total cost of AI) for a workflow: list API prices for T_in/T_out (or a fixed monthly dedicated host), plus review, risk, maintenance, and amortized build. Task-type presets assume agentic AI (retrieval, tools, multi-step) and give a default tokens per task — not an hours-based human comparison.

In production, API inference is often only 15–25% of total AI cost. This worksheet adds maintenance and amortized build so TCA reflects more of the real run cost. Task-type presets are a starting point; for a serious estimate, run a pilot on 50–100 real samples and enter measured tokens below.

TCA = (T_in×P_in) + (T_out×P_out) + (H_review×R_review) + C_error + C_maintenance + C_build

All workload and cost totals are per calendar month. List prices are USD per 1M tokens; ≈ EUR uses ×0.92. H_review is review hours per task; R_review is your review EUR/hr.

Inputs

Task type

Presets assume work is done with agentic AI (tools, retrieval, multi-turn) so API token use stays comparable across task types.

Pilot measurements Use pilot-measured tokens per task Tokens per task (from pilot)

Measure from ~50–100 real samples before trusting presets. When enabled, this replaces the task-type token preset.

Model choice

Volume (units per month)

Task completions per calendar month (tokens and oversight).

Review hourly rate (EUR/hr)

Rate used only for review/oversight cost. Include benefits and overhead if this is base salary only (often 1.25–1.5×).

Review hours per task

Oversight time spent per completed task (not a share of a human task duration).

C_error: risk premium (EUR / month)

Production TCOC_maintenance (EUR / month)One-time build cost (EUR)

Automation rate (% of volume on AI path)

TCA: monthly task cost

Set review hours and C_error under Inputs. Click (?) for formulas and pricing detail.

10 AI units/mo (100% automation) · 3 review h/mo

P_in / P_out (list): $5/1M · $30/1M USD

≈ 100,000 tokens/unit · Monthly T_in / T_out: 300,000 / 700,000

T_in and T_out come from tokens per task × monthly units, with total tokens split ~30% input / ~70% output for costing. Right now: 300,000 T_in / 700,000 T_out per month (from ≈ 100,000 tokens/unit × 10 units). List prices are USD per 1M tokens; EUR uses ×0.92 on the USD API total.

(T_in×P_in) + (T_out×P_out): $22.50≈ 20,70 €
?
(T_in × P_in) + (T_out × P_out) at list USD/1M; approximate EUR applies ×0.92 to the USD total. Using your monthly split: 300,000 T_in and 700,000 T_out → $22.50 (≈ 20,70 €).
H_review × R_review (3 h/mo): 150,00 €
?
H_review × R_review = review hours per task × monthly units × your review EUR/hr (from Inputs). With your inputs: 0.3 h/task × 10 units × 50,00 €/hr ≈ 150,00 €/mo (3 oversight h/mo).
C_error: 0,00 €
?
C_error is the risk / hallucination premium you enter under Inputs (EUR/month). Your value: 0,00 €/mo. It adds to API or fixed cost and oversight to form TCA.
C_maintenance: 0,00 €
?
Ongoing prompt tuning, evals, small fixes, and similar run costs (EUR/month). Your value: 0,00 €/mo.
C_build (amortized): 0,00 €
?
One-time build 0,00 € spread over 24 months = 0,00 €/mo in TCA.
TCA (monthly): 170,70 €
?
TCA = API or fixed infra + H_review × R_review + C_error + C_maintenance + C_build (all per calendar month). With your numbers, total 170,70 € = 20,70 € (API/fixed) + 150,00 € (oversight) + 0,00 € (C_error) + 0,00 € (maintenance) + 0,00 € (build).

Cost summary

Monthly task cost (TCA): 170,70 €
Cost per task: 17,07 €

Worksheet-style check (your numbers, monthly)

Step-by-step recap of the same monthly figures as above: compute, oversight, maintenance, and build. The list below updates live as you change volume, review hours, rate, model, and risk inputs.

AI API: 300,000 T_in + 700,000 T_out ≈ $22.50 (≈ 20,70 € EUR)
Oversight: 3 h × 50,00 €/hr ≈ 150,00 €
Risk premium:0,00 €/mo
Maintenance:0,00 €/mo
Build (amortized):0,00 €/mo
TCA total:170,70 €/mo (17,07 € per task)