AI Model Cost Calculator
Compare live API pricing across Claude, GPT, Gemini, and DeepSeek. Adjust the inputs — the numbers update as you go.
Average tokens per request. Defaults are typical for the selected use case.
The flagship models here cost roughly 25.0× the efficient tier for this workload. Most teams start on an efficient model, measure quality on real traffic, and escalate only the requests that need it — a routing pattern that usually saves 40–70%.
These are back-of-envelope numbers. We build production AI systems — routing, caching, evals, and the infrastructure around them — and can model your actual workload.
They're planning-level estimates based on each provider's own published per-token list price (see the sources linked in the footer). Actual bills depend on your real token counts, any volume discounts you've negotiated, and provider pricing changes — use this to budget, not as an invoice.
Several providers let you reuse parts of a prompt — a system prompt or long context, for example — across requests at a steep discount versus reprocessing it every time. Turn on the caching toggle to see the effect: it applies each model's own cached-input rate to whatever share of input tokens you set with the caching slider.
It models a common cost-optimization pattern: sending most requests to the cheapest budget-tier model in the comparison, and only routing the harder share to a premium model you pick. Production systems often route this way to cut spend without giving up quality on the requests that actually need a stronger model.
The date in the footer shows when these prices were last checked directly against each provider's own pricing page. AI API pricing changes often — if a number looks off, follow the source link next to it.
The comparison spans premium, mid, and budget-tier models from Anthropic, OpenAI, Google, and DeepSeek — see the exact current lineup, tiers, and prices in the table above.
Yes — it's a free tool from Pacifiq Labs. If you want help modeling your actual production workload, routing strategy, or caching setup, use the contact link above.