What is ModelBeat?
One API key and one prepaid balance across every model, with per-request cost accounting. You send auto; the router decides.
ModelBeat is a routing gateway for the multi-model era. You hold one API key and one prepaid balance; for each request ModelBeat decides which model serves the work, executes it against a pool of providers, and returns the response with full per-request token and cost accounting.
It speaks the OpenAI API, so most integrations are a two-line change.
You send auto
model takes auto, or one of three intelligence tiers
when you need a floor on capability rather than on price. You describe the work; the router
picks what serves it. You do not name a model, and the response does not name one back.
Model identity is not in the response body
You do not name a model, and during the current private beta the response body does not
name one back either — model returns the intelligence tier that served the request, not a
model id, and you cannot select, pin, or list a specific model. Route on auto, and treat
the completion as the contract. See Routing
for why this is a beta constraint rather than a permanent one, and Limits —
including the provider-response-header gap tracked as RA-010, which this promise does not yet cover.
The two-line change
from openai import OpenAI
client = OpenAI(
api_key="mb_live_EXAMPLE_NOT_A_REAL_KEY_00000000000000000000",
base_url="https://api.modelbeat.ai/v1",
)
r = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": "hello"}],
)What you get that a provider does not give you
usage.costis what the call cost, broken down by component, in USD. Stock OpenAI does not tell you this. See Costs.- One balance, many providers. You top up once instead of holding a billing relationship with every provider behind the router. See Routing.
X-Request-Idon every response, success or failure, so any call can be traced back to our records. See Errors.
Read this before you build
ModelBeat is in private beta and the published API surface is deliberately narrow:
two metered endpoints, and model: "auto". Limits says exactly what
is and is not supported. Read it first. The constraints will otherwise surprise you late,
when they cost an afternoon.
Next
- Quickstart. A working call in a couple of minutes.
- Authentication. Keys, headers, rotation.
- Capabilities. Tool calling, vision, and the SDKs.
- API reference. The full contract, with a request playground.