One API for every model
Oxyy is a gateway. You integrate once, against an endpoint that speaks the OpenAI API, and reach every model in the catalog through it — with routing, failover and per-request billing handled underneath.
By the numbers
Read live from the gateway, not written by hand. Figures refresh hourly.
The completion figure is the share of all recorded requests that finished without an error, across every model and provider. Requests that failed because of the request itself (an invalid parameter, for example) count as failures too, so it understates how often the gateway itself succeeds. It is not an uptime figure, and we do not publish an availability percentage we cannot measure. Average request duration includes image, audio and video generation, which take far longer than a chat reply. Live provider health is on the status page.
How it works
One endpoint, every model
A single OpenAI-compatible endpoint fronts every model in the catalog. The OpenAI, Anthropic and Gemini SDKs work against it unchanged — you swap a base URL, not a codebase, and switching models is a string change rather than an integration.
Routed, not just proxied
Each model can sit behind several upstream providers. Requests are routed by priority and health, and a provider that starts failing is taken out of rotation and retried elsewhere rather than returning you an error.
Pay for requests, not seats
No subscription and no minimum. Balance is drawn down per request at the rate published on each model page, cache reads are billed at the cached rate, and every charge is itemised in your activity log.
Your keys stay yours
Prompts and completions are not used to train anything. Models that support zero-data-retention or in-region routing are marked as such on their own page, so the guarantee you need is visible before you send the first request.
Start with one request
Create a key, point your SDK at the gateway, and keep the code you already have.