LLM Gateway vs Direct API: What It Really Costs

Calling providers directly has no gateway fee, but it also has no routing, failover or caching layer, unless you build them.

Quick answer

Direct is cheapest only if you don't need routing, failover, budgets or caching across providers. Otherwise, compare the gateway's fee with what it saves.

Direct API: the real costs

List price on every token.
Engineering time for retries, fallbacks, limits and cost tracking.
No multi-provider failover unless you build it.

Gateway: the real costs

The gateway's fee: a markup, a fee on credits, a subscription or a share of savings. See gateway pricing models.
A little added latency, usually small next to model latency. See does a gateway add latency?

The test

Total cost = tokens + fee + engineering time. A gateway that saves 20% on tokens and charges 10% of that saving leaves you well ahead; one that adds a 5% markup without saving anything leaves you behind.

apicuts guarantees your all-in cost, including our fee, is at least 20% lower than going direct. See pricing.

Frequently asked questions

Is a gateway more expensive than calling OpenAI directly?

It depends on the gateway's fee and what it saves. A markup with no savings costs more; a gateway that lowers token costs can cost less in total.

Does a gateway lock me in?

Not if it uses the OpenAI API format: switching back is a base URL change.

Want us to find your 20%?

Book a free 15-minute API cost teardown. Bring last month's provider usage and we'll show you your biggest waste and the first fix we'd ship.

Get a Free 15-Minute API Cost Teardown