Reading a rate multiplier: is 0.15 a 15% discount, or an in-house coefficient?

Why the same 0.15 can differ sevenfold

The same "0.15" on three different sites can mean three different amounts of money:

That is a sevenfold spread. It is not necessarily deception; it is three different pricing languages.

The three-step reading method

Step 1: find the baseline

Open the rate card and look for an "official price" column.

• With a comparison column: the baseline is official pricing, so the published discount can be compared directly.
• Without one: the baseline is almost certainly the site's own number, which may already include a discount or a group coefficient.

Step 2: check the settlement unit

The credit system is where most misreading happens. A site may say "1 yuan = 100,000 credits, multiplier 0.15". The 0.15 applies to credits; you must first find the model's base credit price, apply the multiplier, then convert back to CNY using the top-up ratio. Change any default in that chain and the result moves.

Step 3: back out the unit price

Take the model you actually use, compute the real cost of 1M tokens, and divide by the official price:

effective ratio = (site input + site output) / (official input + official output)

A number near 1.0 means the site prices against official rates. A number around 0.15 means the baseline was already discounted and the 0.15 is a second layer.

Four unit traps

Trap 1: credits, displayed rounded

"Credit display is rounded" means small calls can be rounded up. Negligible per call, meaningful across many short requests.

Trap 2: USD list prices, CNY top-ups

The gap is filled by an exchange rate the site chooses. It may not be 7.1; it may be 7.3, or it may be phrased as "1 yuan = 1 dollar of credit" as a marketing device. Do not read that last phrasing as a real exchange rate.

Trap 3: cache write and cache read have their own multipliers

Of the four prices (input, output, cache write, cache read), cache write is often more expensive than input. If your workload is a long system prompt plus frequent requests, cache lines become a real share of spend, and looking only at input and output will understate cost.

Trap 4: effective multiplier = group multiplier / top-up multiplier

Some sites separate a group multiplier from a top-up multiplier, and the headline effective rate is one divided by the other. You think you are comparing A when you are comparing A / B.

A reproducible worksheet

What to ask the operator

Three questions settle it:

1. "Are your unit prices relative to official rates or to your own baseline?"
2. "How much credit does one yuan buy, and how does that relate to official USD pricing?"
3. "Do the group and top-up multipliers multiply or divide, and which one is used for the final charge?"

A clear answer means the pricing model is theirs and they understand it. A vague answer means they probably have not run the numbers either.

Takeaways

• A multiplier is a relative figure; without a baseline it says nothing.
• When you see 0.15, look for the baseline before you get excited.
• Credit systems, exchange rates, cache pricing and stacked multipliers are the four most common sources of misreading.
• The last step is always to back out the unit price: count money, not numbers.