./tools/cost-calculator — a free tool

Rent the API, or buy the machine? Three years of each.

Type in your team's real numbers — users, queries, tokens — and this tool compares three years of cloud API spend against a one-time hardware purchase plus electricity, then gives you a break-even point in months. Every price is a real published list price with a date and a source, and every field is editable, because your number matters more than ours.

The calculator

Number of users

Queries per user, per day

Input tokens per query

Output tokens per query

API model (published list price, fetched 12 Sep 2026)

These are the vendor's own list prices, fetched 12 September 2026. They are starting points — your negotiated or volume price may differ, so type over them below.

Price per 1M input tokens ($) — editable

Price per 1M output tokens ($) — editable

Hardware being considered

Ballpark starting points below — replace with your real quote.

Hardware price, one-time ($)

Electricity rate ($/kWh)

EIA Feb 2026, New York commercial ≈ 23.5¢/kWh — rounded up to $0.24.

Average power draw (watts)

Assumed average across idle + inference: idle ~15 W, peak ~370 W, blended ~120 W. Assumes 24/7 always-on.

✎ Accuracy note. These are list prices and planning assumptions, not a quote. Cloud prices are list prices per 1M tokens with no volume discount, prompt-caching, or batch discount modeled. The hardware number is whatever you type. Dollar figures are rounded to the nearest dollar (daily and monthly to the nearest $0.10) — directionally accurate, not exact. Scroll to the assumptions list below to audit every number.

Sources (fetched 12 September 2026)

OpenAI API pricing — GPT-4.1, GPT-4o, GPT-4.1-mini, GPT-4.1-nano per-1M-token list prices.

Anthropic (Claude) pricing — Sonnet 5, Haiku 4.5, Opus 4.8 per-1M-token list prices.

U.S. Energy Information Administration — state electricity data — New York commercial average ≈ 23.5¢/kWh (Feb 2026), rounded to $0.24/kWh here.

Hardware prices are not cited because they are your field — the presets are rough round planning figures, not Apple list prices. Verify any figure against a live quote before buying.

Assumptions

Horizon is 3 years = 36 months = 1,095 days.
Electricity: a single flat rate, default $0.24/kWh (EIA Feb 2026 NY commercial ≈ 23.5¢, rounded up).
Hardware runs 24/7; average draw 120 W (idle ~15 W, peak ~370 W under load — Mac Studio-class).
Cloud = list prices only; no volume, caching, or batch discounts, no per-seat software fees.
Hardware = one-time purchase plus power. No financing, IT labor, maintenance, or depreciation — add those yourself if they apply.
Break-even months = hardware price ÷ (monthly cloud cost − monthly electricity). Ignores time value of money.
Token counts per query are a planning guess — your real prompts and outputs will differ.
All dollar figures rounded conservatively; treat as directionally accurate, not exact.

The honest takeaway

This calculator is not rigged toward hardware, because a tool that is rigged loses the sale the moment you check the math. Run a 10-person office at a few thousand queries a day and the API is usually the cheaper, lower-risk option — cloud wins at low volume, highly variable load, or when you'd rather not tie up capital. Hardware earns its keep only when sustained volume is high and steady: enough tokens every single day, for years, that the per-token rent bill outruns a one-time purchase plus power. If the break-even number above lands well inside three years, buying makes sense. If it lands outside — or says "never" — rent, and revisit when the volume is real.

WANT THIS RUN ON YOUR REAL NUMBERS?on-site /packages

The calculator gets you to a direction. If you'd rather have the right machine sized and installed — for your actual workload, not a planning guess — that's the service.

see the packages → get the setup audit or book a call