Separate inference rates from credit fees
OpenRouter's FAQ says it passes through provider inference prices without markup, but charges a fee when you buy credits. A claim that every OpenRouter token has a generic 0–10% markup is misleading. Your effective spend depends on the credit fee, any provider route selected and whether a direct account offers a discount or committed-use terms.
Compare the same model and route
Take a real usage sample with input and output counts. Price it using the exact OpenRouter route and the exact direct provider model. Check whether automatic routing can change provider or model behavior. Do not compare a discounted cached or batch direct rate against a standard routed inference rate without stating the difference.
Operational differences matter
A router may simplify multi-provider access and failover. Direct access may give you earlier features, provider-specific controls, support or account-level agreements. Evaluate rate limits, data handling and outage behavior for your workload. A small token-rate difference can be outweighed by failures or engineering time.
Make a monthly comparison
Use Tokens to USD for the base text estimate, then add credit purchase fees and any non-text or tool charges. Repeat for low, expected and high traffic. Verify live terms before purchasing because pricing and routing options change.