cocodot
← Back to guides
Local card declined for overseas AI? cocodot: one card + one key
PitfallsUpdated 2026-09

Why a '30% Off' Relay Can Cost More Than List: Multipliers, Units and Exchange Rates (2026)

Two relays both say '30% off'; one is cheap, the other costs double list. The gap hides in multiplier bases, pricing units and exchange rates.

TL;DR: '30% off' has no fixed meaning in the relay business. Three common fogs: ① multiplier pricing — some sites show no unit price, only '×1.5', and which base and when it changes is anyone's guess; ② units — some quote per thousand tokens, some per million, so numbers differ 1000x at a glance; ③ exchange rate — CNY-priced sites rarely say which day's rate converts USD models or whether a spread is added. The self-defense: convert every quote to USD per million tokens, input and output separately, then compare — one minute sees through it. The end gives the method and our own convention (listed price = actual charge; savings versus list stated per model).

Trap one: the multiplier's base moves

Multiplier pricing is fine in itself (many platforms use it); the problem is an opaque base. If the base is the vendor's list price, your cost falls when the vendor cuts prices; if the base is the platform's own sticker price, you cannot know in advance when or how much it changes. The test: ask 'how many dollars for one million tokens of Claude Sonnet' — demand a number, not a multiplier.

Trap two: mismatched units, numbers off by 1000x

Chinese models are quoted in yuan per thousand tokens; overseas models in dollars per million. ¥0.015 per 1K looks cheaper than $2 per million, but converts to ¥15 per million ≈ $2.1 — more expensive. Normalize to USD per million tokens before comparing, and always separate input from output: many workloads produce far fewer output tokens, and comparing input alone can invert the conclusion.

Trap three: exchange rate and spread

A CNY-priced platform calling USD models must convert your dollar cost to yuan. Which day's rate, whether a spread (typically 1-3%) is added, whether swings are adjusted retroactively — rarely stated on the page. The test: ask 'is your rate live or fixed, and is there a spread'. If they cannot answer, assume a hidden cost.

A one-minute check of any vendor's real price

① Normalize quotes to USD per million tokens, input and output separately; ② look up the vendor's current list price (the only public baseline); ③ compute the ratio — below list is a discount, above is a markup; ④ test small: top up the minimum, run 10 calls, reconcile each charge against the listed price. Step 4 matters most — the page is a promise, the bill is the fact.

Our convention (use it as a yardstick)

Our pricing page shows, per model, the vendor list price, our price and the percentage saved, in USD, input and output listed separately; time-of-day discounts state the hours — see cocodot.co/pricing. Listed price is the actual charge, failed calls are not billed, and the wallet ledger records every call with model name and token count. You do not have to believe this paragraph — top up a small amount, run it, and reconcile.

Converting three quote styles into comparable prices

What you seeWhat it meansComparable price
×1.5 multiplierBase price × 1.5; base is usually listList × 1.5 = your real cost
¥0.015 / 1K tokensUnit price per thousand tokens× 1000 ÷ exchange rate = USD per million tokens
30% off listList × 0.7 (but ask which list)Look up current list × 0.7

FAQ

How can some sites sell at 30-40% of list?

Legitimate resale discounts come from purchasing volume. Persistently pricing below the vendor's cost is not sustainable — either a subsidy that can stop anytime, or the goods are not what they claim. The test is step 4 above: small top-up plus line-by-line reconciliation.

Is multiplier pricing always bad?

Not necessarily; it is convenient for the platform. Insist on one thing: a concrete figure for 'one million tokens costs how much', and the multiplier becomes transparent on its own.

About cocodot

cocodot is a payment and AI access service for developers and cross-border teams in mainland China. It provides US-BIN virtual cards issued by a licensed institution — used to pay for overseas subscriptions and ad accounts — and an OpenAI-compatible AI API gateway for calling Claude, GPT and Gemini from within mainland China. Both share one wallet, funded by Alipay and accounted in USD. Card: $9.9 to open, 3% to load, $1 per active card per month; spending: $0.60 settlement fee on purchases under $20; a corresponding fee applies when the issuer charges one.

Service scope, pricing and limits →