How Much To Run AI answers the question every team asks right after “should we self-host?”: what would that actually cost? Pick an open-weight model — Kimi, GLM, DeepSeek, Qwen — and it breaks down the budget: GPU purchase versus cloud rental, electricity, operations overhead, and the per-token result next to what the hosted API would have charged you.
This is a calculator, not a platform, and that’s its charm. The self-hosting conversation is usually conducted entirely in vibes — “APIs are expensive”, “GPUs are an investment” — and a tool that replaces vibes with a line-item budget earns its bookmark the first time it ends a meeting early.
What it costs
The calculator is free; accounts save configurations, and there’s a pricing tier above for teams. The irony of paying to learn what things cost is mercifully avoided at the casual tier.
What we can’t tell you
How current its assumptions stay. GPU street prices, cloud spot rates and model efficiency all move monthly, and a cost calculator is only as honest as its most stale number. Sanity-check its GPU prices against a live retailer and its API prices against the provider’s page before taking a result into a budget meeting — if those two spot-checks pass, trust the rest.
The details
| What it is | Cost calculator for self-hosting open LLMs vs using hosted APIs |
| Price | Free calculator · paid tier for saved team configurations |
| Best for | Settling the self-host-or-API argument with numbers instead of vibes |
| Think twice if | Your usage is spiky and small — the API answer wins before you calculate |
