OpenAI Models and Groups

The Codex Relay model list is dynamic. The group assigned to an API key, current upstream account pool, and publication policy together determine which models the key can see and call.

Get the authoritative model list

curl https://codexapi.asia/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY"

Copy the returned id exactly. A client nickname, old configuration, or built-in third-party list is not authoritative.

What a group controls

Area	Effect
API capabilities	Responses, Chat Completions, Embeddings, or Images
Visible models	Model IDs returned by /v1/models
Scheduling pool	Upstream accounts eligible for the request
Billing	Model multipliers, subscriptions, or balance usage
Concurrency and limits	Concurrent requests, RPM, and window limits

Model IDs and aliases

Copy model IDs without modification. Administrators may map a public model ID to an upstream model, but clients should only use the public ID exposed by their group.

If a model is not found

Refresh /v1/models first. If the model is absent, changing the local spelling will not help. Use another group or ask the administrator to publish the model.

Match models and endpoints

Type	Common endpoint	Notes
OpenAI Responses	POST /v1/responses	Preferred by Codex and newer OpenAI clients
OpenAI Chat	POST /v1/chat/completions	Broad compatibility with chat clients
OpenAI Embeddings	POST /v1/embeddings	Vectorization and retrieval workflows
OpenAI Images	POST /v1/images/generations	Requires a group that publishes image models

Capacity and scheduling

One public model can be backed by several upstream accounts. If every compatible account is busy, rate-limited, or unavailable, the gateway may return 429, 503, or Selected model is at capacity.

Keep streaming connections open, honor Retry-After, and use exponential backoff with jitter for retryable failures. Do not retry indefinitely in a tight loop.