API reference

The inference API accepts requests in the OpenAI chat completions shape, the OpenAI Responses shape, and the Anthropic Messages shape, and serves them through the routing engine. Existing SDKs work when pointed at the Inferbase base URL. This section documents the endpoints an application calls, in one shape: method and path, purpose, authentication, parameters, example, response, and errors. Endpoints that serve the portal itself, such as usage reports, receipts and agent runs, are not part of this reference.

Only chat routes. Embeddings, rerank, images, speech, transcription and video are chosen by the routing policy's pool rather than by evidence, because nothing Inferbase measures ranks those models against each other. See Embeddings and rerank and Images, audio and video.

Request shapes and formats

The Inferbase API accepts chat requests in three shapes: OpenAI Chat Completions, the OpenAI Responses API, and Anthropic Messages. It is OpenAI compatible and Anthropic compatible, so the OpenAI and Anthropic SDKs work when pointed at the Inferbase base URL; send the shape your SDK or framework already uses. All three reach the same router, take model: "auto" or a model id, and support streaming and tool calling.

The other surfaces use the OpenAI shape for their kind of request: embeddings, images, speech, transcription and video. Rerank uses the Cohere shape.

Base URLs

Both OpenAI-style base URLs serve the model endpoints with the same authentication, limits, and timeouts. Keys, credits, usage, routing decisions, and model health are served only under https://api.inferbase.ai/api/v1/inference. Use one base URL per integration.

Authentication

Send the key from Gateway › API keys as a bearer token.

Authorization: Bearer inf_your_api_key

Endpoints that create, change, or revoke keys require a signed-in session rather than a key, so a leaked key cannot mint another. Two endpoints need no credential: the model list and a model's health. Routing preview needs a signed-in session. See Authentication and idempotency.

Endpoints

Paths are relative to https://api.inferbase.ai/api/v1/inference. The model endpoints are also served under https://api.inferbase.ai/v1, and Anthropic SDKs reach /messages through their own base.

Quick start

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    api_key="inf_your_api_key",
    base_url="https://api.inferbase.ai/api/v1/inference",
)

# Routed: Inferbase chooses the model from the pool your routing policy allows.
response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Explain quantum computing in 3 sentences."}],
)
print(response.model)
print(response.choices[0].message.content)

# Pinned: a specific model, no routing step.
response = client.chat.completions.create(
    model="openai-gpt-oss-120b",
    messages=[{"role": "user", "content": "Hello!"}],
)

Node.js (OpenAI SDK)

import OpenAI from 'openai'

const client = new OpenAI({
  apiKey: 'inf_your_api_key',
  baseURL: 'https://api.inferbase.ai/api/v1/inference',
})

const response = await client.chat.completions.create({
  model: 'auto',
  messages: [{ role: 'user', content: 'Explain quantum computing in 3 sentences.' }],
})
console.log(response.choices[0].message.content)

curl, streaming

curl -N https://api.inferbase.ai/api/v1/inference/chat/completions \
  -H "Authorization: Bearer inf_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "messages": [{"role": "user", "content": "Hello!"}],
    "stream": true
  }'

Anthropic SDK

from anthropic import Anthropic

client = Anthropic(auth_token="inf_your_api_key", base_url="https://api.inferbase.ai")
message = client.messages.create(
    model="auto",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello!"}],
)

The Anthropic SDK sends its key as x-api-key by default, which the gateway does not read. Pass the key as the auth token, which sends the bearer header.

Conventions

• Model ids are as listed by GET /models: the catalog id of the model, such as openai-gpt-oss-120b, the same id the model's page carries at /models/{id}. A specific variant a route serves is named with a colon, such as deepseek-r1-0528:turbo. The id says which model, never which provider served it. A provider's own spelling of an id, such as openai/gpt-oss-120b, is accepted as an alias on every request; responses always return the catalog id.
• Unrecognized request fields are rejected with a 422, never ignored. A field that is accepted is either forwarded to the provider or reported as not forwarded, so a 200 always means the request you sent is the request that ran.
• The response model is always the model that served the request, never auto. The response id is the request id.
• Errors use the OpenAI error envelope with a request_id. See Errors.

Last updated October 11, 2026.