Savings
Self-host vs API break-even calculator
Monthly token volume at which renting a GPU and serving an open model costs the same as an API model.
Prices read from the vendor pages on 2026-10-20. Arithmetic on list prices, not advice or a quote.
Saving and exports need Norwell Tools Pro, $9 a month. See what is in it. Everything above is free.
How it works
A rented GPU costs the same whether it is busy or idle, so self-hosting is a utilization bet. The break-even is the monthly output volume where the fixed GPU bill equals what the API would charge for the same tokens. Engineering time, model quality and compliance are not in the formula and usually decide it.
The formula
self = gpu_hourly * gpus * 730; tokens = throughput * 3600 * 730 * utilization; api_equiv = output_price + ratio * input_price; break_even = self / api_equiv * 1,000,000Worked example
With the inputs the page opens on:
- gpu hourly cost: 2.5 USD
- gpus: 1
- output tokens per second: 800 tokens
- utilization pct: 35%
- input output ratio: 4
- api model: claude-haiku-5-5
gives
- self host monthly: $1,825
- self host cost per million output: $2.48
- api cost per million output equivalent: $0.900
- break even output tokens per month: 2,027,777,778 tokens
- monthly output tokens at utilization: 735,840,000 tokens
Sources
- https://developers.openai.com/api/docs/pricing (read 2026-10-20)
- https://platform.claude.com/docs/en/about-claude/pricing (read 2026-10-20)
- https://ai.google.dev/gemini-api/docs/pricing (read 2026-10-20)
Prices read from each vendor's own pricing page on 2026-10-20; the page may have changed since. This is arithmetic on published list prices, not advice or a quote. Check the vendor's page before committing a budget.