Models and service providers

In InOneAPI, a model represents the AI capability you call, while a service provider operates the infrastructure that actually processes the request. You use the model ID in your request, and InOneAPI routes it among the available service providers that offer that model.

What a model represents

A model determines what a request can do, such as text conversation, code generation, image generation, speech synthesis, or video generation. Models also differ in input and output modalities, context and output limits, tool calling, structured output, streaming support, speed, and price.

The model catalog and detail page show the currently available capabilities, pricing, and service providers. Verify the capabilities required by your workload instead of inferring them from a model name.

What a service provider represents

A service provider runs the model and processes the request. The same model may be available through several vetted service providers. They offer the same model but can differ in current capacity, response speed, price, and data-processing policies.

One request is not sent to every service provider. Smart routing selects one available service provider and adjusts future traffic as operating conditions change. Multiple providers reduce the risk of an interruption caused by one upstream service, but they do not guarantee uninterrupted service.

Stable model IDs

Requests use the model ID shown by the platform, for example YOUR_MODEL_ID. Switching the service provider normally does not require a model-ID or API change.

Smart routing only selects among service providers for the requested model. Price-first routing favors a lower-priced available provider for that same model; it never silently substitutes another cheaper model.

Choose a model

Before using a model in production, verify its modalities, context limit, tool and streaming support, pricing, available service providers, performance, and relevant data policies. Test representative requests in the playground or a test project before moving production traffic.

Applications should explicitly send the intended model ID and handle timeouts, 429, 5xx, and interrupted streams.

Pricing and request records

The model detail page shows available service providers and their prices. After a request, usage records identify the model and service provider used, together with tokens, cost, latency, retries, and X-Gateway-Trace-ID metadata for billing checks and diagnostics.

InOneAPI does not persist prompt or completion content. Requests must still be sent to the service provider that processes them, so sensitive workloads should also review the provider's data-processing terms.

Next, read Smart routing and Reliability and failover.