Ornn
ServerlessInference

Ornn Inference

Early access

One API for frontier models.

Send a chat completion as a market order on Ornn's token market. Requests take the familiar messages array and responses come back OpenAI-shaped, so your existing code moves over in a few lines.

Request accessSign in

The API

A chat completion is an order

Every request names a model, an order type, and the messages. Authenticate with your Ornn API key.

import os
import httpx

response = httpx.post(
    "https://api.ornn.com/v1/inference/chat/completions",
    headers={"Authorization": f"Bearer {os.environ['ORNN_API_KEY']}"},
    json={
        "model_id": "openai/gpt-4o-mini",
        "order_type": "market",
        "messages": [
            {"role": "user", "content": "Write a haiku about GPUs."},
        ],
    },
    timeout=120,
)
print(response.json()["choices"][0]["message"]["content"])
Response
{
  "id": "chatcmpl-7f3a9c",
  "object": "chat.completion",
  "created": 1792051200,
  "model": "openai/gpt-4o-mini",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Silicon warms up..."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 14,
    "completion_tokens": 17,
    "total_tokens": 31
  }
}
model_id
The model to run, written as maker/model.
order_type
market fills right away at the current price.
messages
The conversation so far: system, user, assistant, and tool turns.
stream
Set to true to receive the completion as server-sent events.
tools
OpenAI-shaped function tools, with tool_choice to steer them.
max_tokens, temperature, top_p, stop
The sampling controls you already use.
order_type: limitComing soon
Set limit_price, the most you'll pay, and the order fills only at or under it.

Pricing

Pay per token

You pay for the input and output tokens each order uses. Prices are per million tokens.

ModelMade byInput / 1M tokensOutput / 1M tokens
GPT-4oopenai/gpt-4oOpenAIComing soonComing soon
GPT-4o miniopenai/gpt-4o-miniOpenAIComing soonComing soon

Open-weight models

Open-weight models are coming

Open-weight models on Ornn Serverless

Coming soon

Open-weight models will be served on Ornn Serverless. Until then, you can run your own open-weight model today with an @app.server unit.

See Serverless