AIVory Smart Inference  ·  Cost check

You pay for tokens every month. See what they cost routed.

Smart Inference sends every request to the cheapest provider that can serve it right now. Put in the models you call and your monthly tokens, and see what the same traffic costs through the router, per model, at today's live prices. No account, no form, nothing leaves your browser.

This works best for open-weight models such as Llama, Qwen, DeepSeek, Gemma and Mistral, because many providers serve the same model and their prices differ. If all your traffic goes to one closed model, there is little to route, and we will tell you so.

Run the cost check How the switch works

Live prices from our own API, on this page.

Cost check

Your traffic, priced on today's routes.

Pick the models you call and how many tokens a month. Put in what you pay per million tokens today, or leave the on-demand list price in. Prices load live from our API; nothing you type leaves your browser.

Loading today's prices…

How the switch works

One line in your code, then the router decides.

There is no new SDK and no migration project. Smart Inference speaks the OpenAI API, so the change is the base URL and the key.

You change the base URL

Point your OpenAI client at https://smart.aivory.net/v1 with an si_ key. Your prompts, your parameters and your streaming code stay as they are.

The router picks per request

For every request it drops the providers that cannot serve it: missing tool calling, a context window too small, or unhealthy right now. Then it picks the cheapest one that is left.

Failures move on

If the chosen provider fails before your stream starts, the next candidate gets the same request. You see the answer, not the outage. One header turns this off if you need a fixed route.

What you can check yourself

Every response tells you what it cost.

You do not have to trust our cost check. After the switch, each response carries its own numbers.

X-SI-Cost

The price of that one request in USD, margin included. Add them up and you have your bill before the invoice does.

X-SI-Model

The model that actually answered. Ask for a tier like light, medium or high and this header tells you which model the tier resolved to.

X-SI-Score and X-SI-Candidates

How many providers could serve the request, and how the winner scored. A high candidate count is where routing saves the most.

X-SI-Spot

Whether reclaimable capacity served the request. Spot is cheaper, and the response says so when it was used.

/v1/pricing

Today's per-model rates, input and output per million tokens, readable by machine. It is the same list our cost check uses.

A monthly cap

Prepaid credits with a monthly spending cap, $100 by default. A runaway loop stops at the cap, not at your card limit.

Set your expectations

We are not a free tier.

Here is the deal in full, both halves of it.

What you do get

What you do not get

What happens next

From the check to the first routed request.

  1. You run the check

    Above, with your own numbers. Prices load live from /v1/pricing, so the result carries today's date.

  2. You sign up and add credit

    Prepaid credits from $10 through Stripe, with a monthly cap. No subscription.

  3. You change the base URL

    Point your OpenAI client at https://smart.aivory.net/v1 with your si_ key.

  4. You watch the bill

    Every response carries X-SI-Cost. Add them up and compare them with the check.

Questions

Before you run the numbers

What does the cost check cost?

Nothing. It runs in your browser against our public pricing API, and the numbers you type never leave the page.

Where does the price I pay today come from?

If you leave it empty, the check uses the on-demand list price we track for that model: the highest on-demand rate among the providers that serve it. If you pay less, type in your own rate and the comparison uses that.

How do you price a request?

Per token, input and output priced separately, at the rates in /v1/pricing. The router buys the request from the cheapest provider that can serve it, and our margin sits between that cost and the price on the list. Every response shows what that request cost you in X-SI-Cost.

Do I have to change my code?

The base URL and the API key. Smart Inference is OpenAI-compatible, so the OpenAI SDKs and most frameworks built on them work with the new URL.

What happens when a provider goes down?

If the provider fails before your stream starts, the router sends the same request to the next candidate, and you never see the failure. Once a stream has started it is locked to that provider. If it breaks mid-stream you get the error, and you pay only for the tokens that arrived.

Can I keep my traffic on one provider?

Yes, per request. Send X-SI-Disable-Failover: true and only the top-scored candidate is tried, ranked on cost alone.

How do I pay?

Prepaid credits through Stripe, from $10. A monthly spending cap protects you, $100 by default, and auto-recharge is optional. There is no subscription.

Which models can be routed?

The live list is GET /v1/models, and the prices are GET /v1/pricing. The model pages show the open-weight models with their current rates.

Run the numbers.

Two minutes with your own token counts tells you whether switching is worth one line of code.

Run the cost check