Token rates pass through; they are not discounted
OpenRouter routes requests to underlying model providers and charges the listed per-million-token rates for the endpoint that serves the request. Prompt and completion tokens can be priced differently, just as they are at the provider.
There is no markup on those inference rates. That means OpenRouter does not lower the provider’s listed token price for a given endpoint. If you compare the same model at the same provider rates, the token line item is a pass-through, not a discount.
Savings, if any, come from other factors: routing to a lower-priced available endpoint for that model, failover behavior, or choosing a different model entirely—not from OpenRouter cutting the base rate.
Credit purchases add a platform fee
Pay-as-you-go users buy credits on OpenRouter. A platform fee of 5.5% applies to those credit purchases. That fee is separate from token pricing.
Because of that fee, the money you put in is not equal one-for-one to the token budget you can spend. Direct provider billing does not use this OpenRouter credit top-up step, so the total customer bill can differ even when the listed token rates match.
Free models are $0 per token, with reliability limits
Free models on OpenRouter cost $0 per token. They are not unlimited. Free accounts are limited to 50 requests per day. After you add $10 or more in credits, the free-model daily limit rises to 1,000 requests per day. Both tiers are also limited to 20 requests per minute.
Failed requests still count against the daily quota. Context windows on free endpoints may also be smaller than paid versions of the same model family. For beginners, that means free access can be useful for light testing, but it is less predictable for steady work.
- Free account free-model limit: 50 requests per day
- After $10+ in credits: 1,000 free-model requests per day
- Rate limit: 20 requests per minute
- Failed free-model requests still consume quota
Routing and lower-cost paid models
OpenRouter can route a model slug to different qualifying providers. Appending :floor to a model slug, or setting provider sort to price, sends the request toward the lowest-cost qualifying provider for that model.
That is routing, not a hidden discount on every provider’s list price. Choosing a lower-cost paid model can also reduce expenses more reliably than relying on free endpoints, because paid usage is not bound by the free daily quotas described above.
Optional tools can change the bill as well. If a request uses extras such as web search, those features can add cost on top of base token charges. Exact tool pricing depends on current product settings and should be checked on OpenRouter’s pricing materials before you assume a total.
One real test, not a price guarantee
In one internal test, Grok 4.3 generated 1,073 output tokens and OpenRouter reported a cost of $0.003602.
That result is a single run. It is not a benchmark for every prompt or a promise of future pricing. Actual cost varies with the prompt, the model, output length, any tools used, the endpoint selected, and current provider rates.
How to compare OpenRouter with direct provider access
Start with the same model and the same expected token mix. Compare the provider’s listed prompt and completion rates with the OpenRouter listing for the endpoint you would actually hit.
Then add the costs that sit outside pure token rates: OpenRouter’s credit-purchase fee if you top up there, free-tier limits if you rely on free models, routing choices such as :floor, and any optional tools.
OpenRouter can still be convenient because one integration reaches many models. Convenience is not the same as a lower list price. For pure cost, treat OpenRouter as a pass-through router plus platform fees and limits—not as an automatic discount on every model.