API ManyNew models: Qwen 3.8 Max, Claude Haiku 5.5, Claude Opus 5.5, Claude Sonnet 5.5

OPENAI-COMPATIBLE API

Keep the SDK. Change the routing layer.

API Many accepts the familiar messages-based chat-completions contract, then separates your logical model ID from the provider-specific route selected behind it.

Compatibility at the boundary—not a promise to clone everything.

The contract covers model discovery, chat completions with streaming and tools, the Responses API, embeddings and transcription. Existing OpenAI client libraries set a different base URL and keep Bearer authentication, JSON messages, usage fields, and the familiar error envelope.

Compatibility does not mean every provider feature is silently translated. Tools, structured output, multimodal content, streaming, caching, and model-specific parameters become available only after their semantics and metering behavior are verified across every eligible route.

That narrow contract is deliberate: it makes failover behavior observable and prevents a provider-specific extension from producing different results on a backup route.

What stays stable

01

Model identity

Your application calls one canonical ID. Provider-specific aliases remain a routing concern.

02

Request receipt

Public model, request status, usage, and final charge share one API Many request ID. Internal routing details are excluded.

03

Failure meaning

Definite upstream rejection may fail over; ambiguous acceptance stops for reconciliation.

ONE PUBLIC CONTRACT

Start with an OpenAI-compatible request.