Private beta

One API key.
One balance.
Every model.

ModelBeat routes each request to the model that should serve it, then hands back the response with the served model and the exact cost attached. It speaks the OpenAI API, so most integrations are a two-line change.

QuickstartAPI reference

Two metered endpoints, streaming supported. Read the limits first.

pythonquickstart
1from openai import OpenAI
2
3client = OpenAI(
4    api_key=os.environ["MODELBEAT_API_KEY"],
5    base_url="https://api.modelbeat.ai/v1",
6)
7
8r = client.chat.completions.create(
9    model="claude-haiku-4-5",
10    messages=[{"role": "user", "content": "hello"}],
11)
12
13# which model actually served it, and what it cost
14print(r.extra_fields["routing_info"]["model"])
15print(r.usage.cost["total_cost"])

What comes back with every call

routing_infoThe served modelWhich model actually answered, whether it was a fallback, and what was originally targeted.Read →usage.costWhat it costPer-request USD, broken down by component. Reserved before routing, settled after.Read →X-Request-IdA thread to pullOn success and on failure. Quote it and we can find the exact call.Read →