MEGA Hub
New
Docs/Get started

Your First Inference Provider Call#

This guide takes you from a live model mapping to a successful routed request. MEGA runs the model on a healthy registered Provider; you do not deploy an endpoint first.

Prerequisites#

You need a MEGA account and a Personal Access Token (PAT) with the inference:run scope. Create the token under Settings → Access Tokens, copy it once, and keep it outside source control:

BASH
export MEGA_TOKEN="mega_..."

Step 1: find a live model#

In the browser, open the Inference Provider model comparison to filter live model/Provider routes and compare price, context length, latency, throughput, tool calling, and structured-output support.

The same live catalog is available through HTTP:

BASH
curl https://inference.tensorplay.cn/v1/models \  -H "Authorization: Bearer $MEGA_TOKEN"

Use the public Hub catalog when you need MEGA-specific Provider, price, and capability fields:

BASH
curl https://mega.tensorplay.cn/api/inference/models

Each model contains its live Provider mappings. Prices are USD per one million tokens, and one model can expose multiple Providers or tasks.

Step 2: call with the OpenAI SDK#

Install the OpenAI SDK:

BASH
python -m pip install openai
PYTHON
import osfrom openai import OpenAI client = OpenAI(    base_url="https://inference.tensorplay.cn/v1",    api_key=os.environ["MEGA_TOKEN"],) response = client.responses.create(    model="mega/gpt-5.4-mini",    input="Reply with exactly: MEGA is ready",)print(response.output_text)

Chat Completions and Embeddings use their normal OpenAI SDK methods:

PYTHON
chat = client.chat.completions.create(    model="mega/gpt-5.4-mini",    messages=[{"role": "user", "content": "Hello"}],) vectors = client.embeddings.create(    model="BAAI/bge-m3",    input=["first input", "second input"],)

Step 3: call with raw HTTP#

BASH
curl -i https://inference.tensorplay.cn/v1/chat/completions \  -H "Authorization: Bearer $MEGA_TOKEN" \  -H "Content-Type: application/json" \  --data '{    "model": "mega/gpt-5.4-mini",    "messages": [{"role": "user", "content": "Reply with exactly: MEGA is ready"}]  }'

Every success, error, and streaming response includes an Inference-Id header. Record it when reporting a failed call; MEGA does not need the prompt or response content to trace routing and billing state.

Choose a Provider#

Provider selection is encoded in the API's model field and works identically in OpenAI-compatible SDKs and raw HTTP:

Model value Behavior
owner/model Fastest healthy Provider; the default.
owner/model:fastest Explicit fastest selection.
owner/model:cheapest Lowest standard output-token price, then latency.
owner/model:preferred First healthy Provider in your account or organization preference order.
owner/model:provider-slug Restrict the request to one Provider.

For example, send mega/gpt-5.4-mini:cheapest or mega/gpt-5.4-mini:mega as the model value.

MEGA can fail over only before response headers or the first streaming byte are sent. It never transparently retries after content starts, preventing duplicated output and duplicated settlement.

Routed billing, custom keys, and organizations#

The default auto billing mode follows the configured account: when the selected Provider has a saved custom key, MEGA swaps it in and records zero MEGA inference cost; otherwise MEGA preauthorizes Routed Inference and settles the Provider's standard price without a service markup. Your request code does not change after adding or rotating a saved key.

After you save an encrypted Provider key in Settings → Inference Providers, normal auto calls use it automatically when that Provider is selected. To require a saved key and fail instead of falling back to routed billing, set a request header:

PYTHON
byok = OpenAI(    base_url="https://inference.tensorplay.cn/v1",    api_key=os.environ["MEGA_TOKEN"],    default_headers={"X-Mega-Inference-Billing": "byok"},)response = byok.responses.create(    model="mega/gpt-5.4-mini:groq",    input="Hello",)

MEGA authenticates the PAT, decrypts the saved Provider key only for dispatch, and records zero MEGA inference cost. The Provider bills the key owner directly. The key is never returned to the SDK or browser.

Set X-Mega-Inference-Billing: routed to ignore a saved key for one call. Set it to byok to require a saved key and fail instead of falling back to MEGA billing.

Eligible Team and Enterprise members with write or admin access can bill an organization:

X-Mega-Bill-To: research-lab

Omit X-Mega-Inference-Billing, or set it to auto, for the normal automatic key-swap behavior.

Continue with Routing and Provider Selection, Pricing and Billing, or the Router API Reference. For failures, keep the Inference-Id and follow Troubleshooting.