Demo · interactive estimator · not a quote

API vs self-hosted: what will your LLM bill be?

Plug in your expected volume and see an estimated monthly API cost next to an estimated self-hosted cost, plus the breakeven volume where the two cross. All math runs in your browser from your inputs — nothing is hard-coded.

Demo · interactive · runs in your browser

LLM cost calculator

Adjust the inputs. Every number below is computed from them live.

Estimated monthly API cost
$4.50
Estimated monthly self-hosted cost
$648.00
Breakeven: requests / day
72,000
Above this, self-hosting is estimated cheaper; below it, the API is.

Estimates only — computed from your inputs and the stated assumptions below. Not a quote, and not a recommendation to self-host.

Assumptions — the exact numbers used

The calculator applies these assumption values to your inputs. They are plausible mid-range illustrations, not live market prices:

Assumption8B class70B class
API price, per 1M input tokens$0.10$0.60
API price, per 1M output tokens$0.20$1.20
Self-host GPU rate$0.90/hr (one 24 GB GPU)$3.50/hr (two 80 GB GPUs)
Self-host utilization mathDedicated instance, 24/7: hourly rate × 24 hours × 30 days. No engineering, monitoring, or failover costs included.

How breakeven is computed: monthly API cost grows with every token, while the self-hosted cost is fixed for the month. Breakeven requests/day = monthly self-hosted cost ÷ (30 × API cost per request). Below the breakeven volume the pay-per-token API is estimated cheaper; above it the fixed self-hosted capacity is.

Why this demo exists: founders routinely over- or under-estimate inference spend by 10×. A two-minute estimate with stated assumptions beats a guess — and it is the same arithmetic we walk through in the free assessment before recommending hosted vs self-hosted.

Frequently asked questions

Are the prices on this page real market prices?

No — every price on this page is a stated assumption used to make the arithmetic transparent. API prices move, GPU rental rates move, and your negotiated rates will differ. Treat the output as an estimate for planning conversations, then plug in your own numbers with your provider.

When does self-hosting usually win?

At high, steady volume. The self-hosted cost is mostly fixed per month, while API cost grows with every token. The breakeven number below is exactly the volume where the two lines cross.

Why is the self-hosted cost flat on this calculator?

The model assumes a dedicated instance running 24/7 for the month — that is how most teams rent GPUs for always-on inference. If your traffic is bursty, serverless or scale-to-zero hosting can sit between the two.

What does this calculator leave out?

Engineering time, monitoring, evals, red-teaming, failover, and the cost of a bad answer reaching a customer. Those are real costs of self-hosting — the calculator covers the compute line only, so bring it to the free assessment and we'll walk through the rest.