Models
List available models with GET /v1/models and see the OpenRelay public model ids you can pass as the model field in an inference request.
Models are addressed by their public OpenRelay id. Use a model id in the
model field of a chat completion, a
message, an embeddings
request, or a decisions request.
List models
GET https://api.openrelay.inc/v1/modelsRequires an API key (see Authentication). This
endpoint is free: it is not balance-gated. Model discovery is served by the
control plane (api.openrelay.inc); inference itself runs on
inference.openrelay.inc.
curl https://api.openrelay.inc/v1/models \
-H "Authorization: Bearer $OPENRELAY_API_KEY"The response is an OpenAI-style list object:
{
"object": "list",
"data": [
{
"id": "openrelay/gpt-oss-120b",
"object": "model",
"created": 1767322800,
"owned_by": "openrelay",
"supported_apis": ["chat_completions"],
"endpoints": { "chat_completions": "/v1/chat/completions" },
"modalities": { "input": ["text"], "output": ["text"] },
"status": "available"
},
{
"id": "openrelay/flux-image",
"object": "model",
"created": 1767322800,
"owned_by": "openrelay",
"supported_apis": ["images"],
"modalities": { "input": ["text"], "output": ["image"] },
"status": "request_access"
}
]
}| Field | Description |
|---|---|
id | The public model id to put in model. |
created | Unix timestamp (seconds) the model was added to the catalog. |
owned_by | Always openrelay. |
supported_apis | The API families this model serves (e.g. chat_completions, messages, embeddings, decisions). Check this field to see whether a model supports Messages; not every model does. |
endpoints | The path to call for each family in supported_apis, keyed by family name, e.g. { "decisions": "/v1/decisions" }. Paths are on inference.openrelay.inc. Only families the gateway serves appear, so a client can call a family it does not know from this map; the field is omitted when none of a model's families has an endpoint, as with images above. |
modalities | What the model accepts and produces: { "input": [...], "output": [...] }. Values are free-form (text, image, audio, video, …), so a text→image, image→image, vision (image→text), or video/world model is expressed without any special-casing. |
status | available (usable now) or request_access (listed for discovery but not yet generally available). |
request_access models appear in the catalog so you can discover them, but a
chat request to one returns a 403 model_access_required error until the model
is enabled. Contact us to request access. Only
status: "available" models can be invoked.
Always use the model id exactly as returned by GET /v1/models. It is the
same id that appears in every response's model field.
Messages-capable models
openrelay/glm-5.2 is the current example of a model whose supported_apis
includes messages, so it can be requested through either
Chat Completions or
Messages on the same model id.
Available models
openrelay/deepseek-v4.1-flash-1017-heretic serves the community-modified
pqhaz Heretic checkpoint.
| Model id | Description |
|---|---|
openrelay/gpt-oss-120b | A large open-weight general-purpose chat model: the higher-capability option. |
openrelay/gpt-oss-20b | A smaller, faster, lower-cost open-weight chat model. |
openrelay/gemma-4-31b | Google's Gemma 4 31B: vision (text + image input) with reasoning and tool calling, 256K context. |
openrelay/glm-5.2 | Zhipu's frontier GLM 5.2: bilingual reasoning and agentic tool use, 1M-token context. |
openrelay/glm-5.3-flash | Zhipu's GLM 5.3 Flash: natively multimodal (text + image input) with reasoning and tool calling, 256K context. |
openrelay/deepseek-v4.1-flash-1017-heretic | pqhaz Heretic checkpoint of DeepSeek V4.1-Flash-1017: text and image input, reasoning and tool calling, FP8 on MI355X, 393,216-token shared prompt and completion budget. Chat Completions only. |
openrelay/deepseek-ocr-2 | DeepSeek's document OCR model: converts scans, receipts, and screenshots to structured markdown (text + image input), 8K context. |
openrelay/bge-m3 | BAAI's BGE-M3 multilingual embedding model: 1024-dimension vectors, 100+ languages, 8192 tokens per input. Embeddings only. |
openrelay/laya | Laya, an encoder-only typed-decision model: choice, score, and yes/no answers with probabilities, English and multilingual checkpoints. Decisions only. |
Call GET /v1/models for the authoritative, up-to-date list available to your
organization. For per-model rates, see Pricing & billing.