One API for every model. Your OpenAI code already works with it.

Change the base URL and keep everything else. Each request goes to the cheapest provider that is up, at the price that provider charges. Tokens bill at the rate the vendor invoices us; the platform fee is charged once, when you buy credits.

Your client does not change. The line under it does.

Point the OpenAI SDK you already use at https://api.blazerail.com/v1 and keep the same model ids, the same request bodies and the same response shape. The gateway decides which provider serves each request; your code never learns there was a choice.

Say it's two in the morning and a provider starts returning errors.

The gateway looks up every upstream serving the exact model id you asked for and orders them by the price it is invoiced and the latency it has measured. The cheapest healthy one is tried first. If it returns an error, the same request is retried on the next healthy upstream serving the same model id; it is never swapped for a different model. The failure is recorded against the upstream that caused it and feeds the next ranking.

Same model, same tokens, a smaller invoice.

Both price lists are read live and matched by model id. Where the provider's own rate is the price, BlazeRail and OpenRouter show the same number. Where a cheaper upstream serves the model here, the page shows the gap per model, and the full comparison lists the models where OpenRouter is cheaper too.

Monday you told it your stack is Rust. Saturday, on a different model, it still knows.

Every conversation through the gateway can leave facts behind: what was said, which conversation it came from, and which earlier fact it replaced. Any model you call later can read them. Retrieval runs over vector, full-text and knowledge graph indexes. Document extraction is $0.01 per PDF page and $0.005 per image.

Every model, its publisher, and what it costs right now.

The numbers on the page are read from the same catalogue and the same measurements the gateway routes on. If a model has had no traffic in the window it is left out, not estimated.

What the alternatives ship, side by side.

Checked against each vendor's published documentation. LiteLLM is the right choice if you want to host the gateway yourself, and OpenRouter if routing is all you need. BlazeRail bundles memory, document extraction, response caching and provider failover behind the same OpenAI-compatible endpoint.