Picking the best model shouldn't mean defending the bill.
Inferray is an inference API: one consistent interface to the models teams actually want, with no gateway fees on top of what you use.
Why we built it
Ship anything real on a flagship model and inference stops being a rounding error. So teams start trimming: shorter context, fewer retries, a weaker model on the paths nobody is watching. Every one of those cuts makes the product a little worse, to claw back money that was never the fair price in the first place.
You pay the rate the model is listed at, with no gateway fees on top — so the model you would have picked on merit is the one you can afford to keep.
The cheaper option and the better option should be the same option.
What that means in practice
No asterisks, and nothing that only applies once you are big enough.
- One interface for every model
- Models from every lab we carry answer behind a single endpoint and a single key. Your code speaks the interface it already speaks — Chat Completions, Messages, or Responses — and streaming works the way it always has.
- No gateway fees
- Nothing is added on top of the rate the model is listed at. There is no trial window, no volume tier to reach, and no plan that unlocks it — the price you see applies to your first request and every one after it.
- The same model, not a substitute
- Same model, same version, same answers. Nothing is quietly swapped for something cheaper, and the model you name in the request is the model that answers it.
- You pay for usage and nothing else
- No seats, no monthly minimum, and no support tier. Top up a balance and spend it down, or pick a Token Plan for one flat monthly price with 3× its value in usage; when the credit runs out, requests stop rather than silently overspending.
- Your prompts stay yours
- We don't store your prompts or completions, and we don't train on them. We keep the token counts we need to bill you; the model provider that serves a request handles it under its own data policy.
- You can see where it went
- Every request shows up in your console with latency, tokens, and cost broken out by project, key, and model — so the bill is never a single unexplained number.
Two lines of config, then the same code
Swap your base URL and your key. Your SDK, prompts, tools, and agents stay exactly as they are.