Frontier models, through the same API.
Models that cannot run inside the Kingdom — served by named providers abroad, or on Sigmix GPUs abroad. Same request shape, same keys page, same credit. Not for data that must stay in the Kingdom: use the sovereign route for that.
https://inference.sigmix.sa/international-route-only/v1Authenticate with a key that starts sgx-intl-. Keys for the other route are refused here.
What this route is, plainly
Served abroad
Requests may be answered outside the Kingdom, by the providers named on the model list. Treat it as you would calling those providers directly.
Never mixed
International keys (sgx-intl-…) cannot reach the sovereign route, and sovereign traffic never falls back here.
Same billing
Reserved before it runs, settled to the exact usage, from the same prepaid balance in dollars or riyals.
We do not store prompts or outputs. Providers on this route process requests under their own terms; the model list names who they are.
Your first call
Any OpenAI SDK works: set the base URL and the key. The samples use the first chat model published on this route.
No models are published on this route yet. They appear here, with samples, as soon as they are.
Models and prices on this route
The live roster. Use the model id exactly as shown.
Showing SAR. The switch in the navigation changes it.
Endpoints
| Endpoint | What | Billed | On this route |
|---|---|---|---|
GET /international-route-only/v1/models | List models | Free | Available |
POST /international-route-only/v1/chat/completions | Chat | per 1M tokens, input · output | No model yet |
POST /international-route-only/v1/embeddings | Embeddings | per 1M input tokens | No model yet |
POST /international-route-only/v1/audio/speech | Text to speech | per 1M characters | No model yet |
POST /international-route-only/v1/audio/transcriptions | Speech to text | per hour of audio | No model yet |
POST /international-route-only/v1/audio/translations | Speech to text | per hour of audio | No model yet |
POST /international-route-only/v1/images/generations | Image generation | per image | No model yet |
Rules that apply on every request
Billing
The most a request could cost is reserved before it runs and settled to the real usage after. A non-streamed response carries x-sigmix-charge-usd with the exact charge; a stream's charge appears on the Usage page when it ends.
Streaming
Available on chat models above an available balance of $20.00. A stream you close early is charged only for what was generated.
Limits
60 requests a minute per key unless you set your own; per-key concurrency and spend caps are yours to set in the console.
Every error has a code that says whose it is. The full list