▸ Live calculator · No signup

AI Infra Cost Calculator

Tokens × volume × model mix + compute + storage. Defaults sourced from real client data, November 2025. Tweak anything.

Monthly user requests

e.g. 200,000 = ~6,500/day

Avg input tokens / req

Avg output tokens / req

Cache hit rate

35% — typical with prompt caching for repeat-system-prompts

Frontier model share (Claude Opus / GPT-class)

20% of requests · ~$15 input / $75 output per Mtok

Mid-tier share (Sonnet / GPT-mini)

55% of requests · ~$3 input / $15 output per Mtok

Small model share (Haiku / open weights)

25% of requests · ~$0.80 input / $4 output per Mtok

Vector DB rows (millions)

Compute baseline ($/mo)

Compute baseline = your orchestration, gateways, observability stack.

Estimated monthly run cost

Excludes engineering time. Includes inference, caching savings, vector DB, and compute baseline.

Frontier model spend$0

Mid-tier spend$0

Small model spend$0

Vector DB$0

Compute / observability$0

Cache savings−$0

Heads up: Model mix percentages should sum to 100. We rebalance proportionally.

These prices change. As of Nov 2025 we use vendor list prices, not committed-spend discounts. For a cost-tuned architecture, talk to us.