Savings Calculator

What would a gateway save you?

Answer four quick questions about your workload — we'll estimate savings from cross-provider arbitrage, caching, and cheap-model routing. No usage data needed; the math runs in your browser against typical workloads in each category.

Your scenario

What are you building?

How big is your user base?

How chatty are your users?

Which model tier do you want?

Estimated monthly savings

On $49 of LLM spend across roughly 180,000 requests over 30 days:

You'd save

$8 – $16

16% – 33% (conservative – aggressive)

Your bill becomes

$32 – $41

after the 4-strategy stack

Estimates based on average workloads in each category. Your actual numbers depend on prompt patterns + cache reuse — drop your real usage into the detailed calculator below for exact math.

Where the savings come from

Cheap-model routing

Routes a fraction of flagship-model traffic to the cheap-fast model in the same provider family.

$0 – $0

medium confidence

Cross-provider arbitrage

Routes a fraction of in-tier traffic to the cheapest provider with the same quality tier.

$4 – $7

medium confidence

Prompt cache

Enables provider-native prompt caching on the cacheable portion of input tokens.

$2 – $7

high confidence

Semantic cache

Quality-validated semantic cache serves a fraction of requests without hitting the provider.

$2 – $11

medium confidence

Now factor in the platform

The savings above are the reduction in provider cost. What you actually pay per month depends on which Leanroute plan you're on. For an underlying spend of $37:

Pay-as-you-go credits5% fee applied at top-up

$39

BYOK Starter (breaks even above $300/mo underlying) and Pro (breaks even above $500/mo) hidden at this volume — the flat monthly fee would exceed any per-request markup you'd save on this workload. Credits is the fair pick until your spend grows.

OpenRouter (reference)5% per-request fee

$39

Have real usage data?

Upload an OpenAI or Anthropic export CSV (or paste a summary) and we'll compute savings against your actual workload — same engine, real numbers instead of scenario averages.

Detailed calculator