A smaller AI bill without rewriting your code

  • Every model behind one key
  • Simple requests 20× cheaper
  • Hundreds of billions of tokens a month

The router picks the best-value model for every request by price, speed and quality. One endpoint with fallback hosts, a log of every request and spend controls.

The same simple request

All on the strongest model100%
With Auto5%
20×cheaper

Built for engineers.
Loved by finance.

Engineers get the best model for every task. Finance gets a smaller bill.

How it works →

Where the savings are

98% of requests don't need the most expensive model

The router pays for a strong model only where nothing else will do.

1.57%

of requests needed the most expensive model

306,180 on cheaper models4,880 on the most expensive

311,060 model calls in the Samreshuuu app over the 14 days to 11/18/2026.

Savings on autopilot.

New ways to save join the router as soon as they prove themselves. Your integration stays the same.

Smarter defaults.

We test new models on real work. The best one becomes your default.

Every model. One endpoint.

Text, images, video and speech from vetted hosts. If a host goes down, the request moves to the next one.

Spend control

AI spend under control from request to invoice

AI works in milliseconds, but spend shows up at month end. We bring the two together: spend is visible on the day of the call, and the statement arrives for the month.

Monthly cap

The balance never spends more than the amount you set.

Auto top-up

The balance tops itself up, so your product keeps running.

Request log

The model, time and price of every call.

Invoice and statement

Top up by bank transfer, get a statement for the month.

How we pick models

We pick models by independent benchmarks and our own tests

Every model is checked against independent benchmarks and our internal tests. The router gets the one that wins on quality, speed and price.

Quality
an overall task index: code, reasoning, knowledge
Speed
how long the model takes to finish the task
Price
what the whole solved task costs

Why not just the strongest one

The top model is often several times pricier and slower, with almost the same quality. It is worth paying for only where the difference shows.

Running at home

Proven on Samreshuuu

The Samreshuuu app runs on this router every day. Everything we learn in production goes straight into the product.

336B
tokens in 30 days
443K
model calls in 30 days
98%
of requests without the most expensive model

Model calls of the Samreshuuu app, as of 11/18/2026.

Setup

Two lines of code

A new base URL and "auto" instead of a model name. The rest of your code stays the same.

python
import os from openai import OpenAI client = OpenAI(api_key=os.environ["SAMRESHUUU_API_KEY"], base_url="https://api.samreshuuu.ru/v1") reply = client.chat.completions.create( model="auto", messages=[{"role": "user", "content": "Hi! What can you do?"}], ) print(reply.choices[0].message.content)
  • No subscriptionPay only for calls, from your balance.
  • Coding agentsOne setting, spend in the same log.
  • DocumentationExamples in cURL, Python and Node. Open →

Questions

Questions and answers

Top up the organization balance by invoice from a company account or by card. A statement arrives for each month.

No. You pay only for calls, with no minimum payment.

A simple request goes to a simpler model, a hard one goes to a strong one. If you need the same model every time, name it.

We don't write request or response text to the log: it holds the model, host, tokens and price.

A developer changes the base URL and the model name. Everything else stays as it was.

Tokens are money. Save both.