Selected models are free to useView free models
Start free

How to Check an LLM API Relay Serves the Real Model

By LanoX TeamPublished Updated Data as of

No single request can prove which model answered. What you can do is run checks whose results you verify yourself: compare token counts with the official counter, reconcile every charge against the published price, and watch whether the price list stays stable. This guide describes those methods. It does not report test results, because LanoX has not published any.

What this guide does and does not claim

LanoX has not published an in-house or third-party test that proves model identity. Everything below is either a method you can run with your own key and prompts, or public price data you can check line by line.

Check 1: compare token counts with the official counter

Anthropic offers a token counting endpoint that is free to use, within rate limits. Send the same prompt to it and, through the relay, to the same model ID. Then compare the input token count the relay reports in its usage data.

Anthropic notes that its count is an estimate and can differ slightly from the tokens actually billed, so look for large gaps that repeat across prompts, not a handful of tokens. It also notes that Claude Opus 4.7 and later models use a newer tokenizer, so do not compare against counts taken on an older model.

Check 2: reconcile the charge against the published price

Without caching and within one price tier, charge = input tokens ÷ 1,000,000 × input price + output tokens ÷ 1,000,000 × output price. That is the formula on the LanoX pricing page FAQ.

Example with the LanoX price for claude-opus-5 ($1.495 input, $7.475 output per 1M tokens): 20,000 input and 4,000 output tokens = $0.0299 + $0.0299 = $0.0598. If the charge in the console differs, ask why.

The LanoX docs say the console shows the requested model, the actual result, the usage record and the final charge, so you can match each line against your own request log.

Check 3: use the model list your key can reach

GET /v1/models lists the models available to your API key, as described in the LanoX docs. Use an exact ID from that list in your requests and keep it in your own logs, so every record maps to one model you chose.

Check 4: watch whether prices stay stable

A price that changes without notice makes reconciliation impossible. We record the LanoX price list every day. For the 10 Claude models on our Claude API pricing page, 16 days of records show 0 price changes (Claude Sonnet 5.5 has 10 days, from its first listing on 2026-10-24). On 2026-11-02 the official prices LanoX lists for them matched Anthropic’s pricing page.

On that date the LanoX price for these models was 40.0% to 70.1% below Anthropic’s list price, worked out per model on the same page.

What price data can and cannot tell you

Price alone cannot show which model answered. Our price gap index also shows LanoX is not the cheapest on every model. Use price data to reconcile bills; use token counts and request records to check the model.

FAQ

Has LanoX published a test proving its models are real?
No. LanoX has not published such a test. This page only describes checks you can run and public data you can verify.
Do I need an Anthropic account for the token count check?
Yes. The token counting endpoint is part of Anthropic’s API, so you call it with your own Anthropic API key.
Where do I see the record for each call on LanoX?
In the console. According to the LanoX docs it shows the requested model, the actual result, the usage record and the final charge.

Sources