One API. Blazing fast inference. Auto-fallback built in.
Change one line of code and never worry about model outages, slow responses, or runaway token bills again. Just call model="smart"—we read each request, judge how hard it is, and route to the right model automatically.
# Drop-in OpenAI SDK replacement
from openai import OpenAI
client = OpenAI(
base_url="https://api.unifiedllm.io/v1",
api_key="sk-uf-..."
)
res = client.chat.completions.create(
model="smart",
messages=[{"role": "user", "content": "..."}],
extra_body={
"routing": {"prefer": "cheapest"}
}
)Describe your task. See exactly what runs.
Preview and live production requests use the exact same routing rules—what you see here is what answers your API call.
The router reads the prompt (length, code, reasoning, tools, images) to judge how hard the task is. Nothing is sent to a model.
from openai import OpenAI
client = OpenAI(base_url="https://api.your-domain.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="smart",
messages=[{"role": "user", "content": "Your prompt here"}],
)Switch in one line
Keep your existing OpenAI SDK, LangChain, or LlamaIndex code. Just change base_url and your API key.
const client = new OpenAI({
baseURL: "https://api.openai.com/v1",
apiKey: process.env.OPENAI_API_KEY,
});const client = new OpenAI({
baseURL: "https://api.unifiedllm.io/v1",
apiKey: process.env.UNIFIED_API_KEY,
});Pay only for what you use. No monthly fees.
Start with as little as $1 USD via Visa / Mastercard. Every response includes exact token count and USD charge. Payment trouble? Email support@unifiedllm.io and we will try to accept your local currency.
Frequently Asked Questions
How do I know which model actually answered my request?
We never hide the answering model. Every API response returns the exact model slug, server cluster, input/output tokens, speed (TPS), USD charged, and whether a backup model was used.
How am I charged if the primary model fails and a backup answers?
Failed attempts are never charged. You are charged strictly at the per-token rate of the model and server that actually delivered your answer.
How is speed (tokens/sec) measured?
Speed numbers on our site come directly from measured p50 production responses across our server clusters, alongside measured baseline speeds of standard providers.
Do you store my prompts and completions?
No—by default we store zero prompt or completion text. You can optionally enable 7-day prompt logging in Dashboard → Settings for debugging.
Can I pay in my local currency or if my payment fails?
Your wallet balance and all token prices are in USD, and top-ups are by card. If you have any trouble paying or need another payment method, email support@unifiedllm.io and we will try to accept payment in your local currency or resolve the issue.
What happens if all 3 fallback attempts fail?
Your pre-authorized balance hold is released 100% and you receive a clean OpenAI-compatible error with zero charge.