What an AI gateway does

An AI gateway sits between your application and the model provider. Your code sends a request to one endpoint, and the gateway handles authentication, model names, routing, fallbacks and usage records. A team can change the model behind a task with a smaller integration change than maintaining a separate connection for every provider.

A gateway and a router often appear together. The gateway is the common access point. The router chooses which model or provider handles a request according to rules such as availability, price, latency or data policy.

How OpenRouter charges

OpenRouter says model inference is billed at the provider's posted list price. Its Standard plan adds a 5.5% fee when credits are purchased. Bring-your-own-key use has a stated monthly allowance, followed by a usage fee above that allowance. Those terms were checked on 5 November 2026 and can change with the provider page.

This means a token price and a final account charge answer different questions. Compare AI Cost records the dated model price. A purchase decision should also include the current platform fee, tax and any direct-provider discount.

When a gateway can reduce the bill

  • Model choice: route routine classification, extraction or summarisation to a capable lower-cost model.
  • Provider choice: use comparable endpoints for the same model where routing terms support a lower price.
  • Fallback control: pay for the successful route while improving availability for a production workload.
  • One usage record: review spend across models in a shared activity log and apply budgets in one place.

When a direct API may cost less

A direct provider account can be the cheaper route when one model handles nearly all traffic, the provider offers committed-use or enterprise pricing, or the gateway fee is larger than the routing saving. Direct contracts can also carry specific data, support and regional terms that matter more than the token rate.

A practical comparison

  1. Measure monthly input, output, cache and batch tokens.
  2. Price the same workload against the exact model versions.
  3. Add the gateway fee, tax and any provider discount.
  4. Check rate limits, context tiers and data-routing needs.
  5. Run a small production-shaped test before moving traffic.

The cheapest route depends on the workload. The useful decision is the complete monthly cost for the quality, reliability and control the task needs.

Sources and next steps

OpenRouter product and fee details come from its official pricing and model pages, checked on 5 November 2026. Current model price records are available in the Compare AI Cost table.