Relay API
Combine two-part pricing into one number
Relays rarely quote a price per million tokens. They give you a top-up ratio and a model multiplier — multiply them to get what you actually pay. This calculator does that multiplication and nothing else; it uses no third-party quotes.
Leave the multiplier at 1 if you don't know it — see the direct cost first, then plug in the multiplier you were quoted.
After conversion, per 1M tokens
Input / output, CNY
- Going direct
- —
- Saving vs official
- —
- The maths
- —
One line of formula
Real CNY cost = (official price × model multiplier) ÷ credit per ¥1
A concrete example: DeepSeek flash off-peak is $0.15 input, $0.60 output.
Say a relay offers "¥1 = $1 credit, multiplier 0.3". Input cost = 0.15 × 0.3 ÷ 1 = $0.045, which is 30% of the official price.
But if it says "¥1 = $0.2 credit", the same 0.3 multiplier gives 0.045 ÷ 0.2 = $0.225, which is 50% more expensive than going direct.
That's why "0.3 multiplier" on its own means nothing. The top-up ratio is the denominator — a lower multiplier doesn't mean cheaper.
Three common traps
- Different multipliers for input and output.Some relays discount input only and bill output at full rate — a big difference for long answers.
- Multipliers are grouped by model.On one relay it's normal to see 1.5× on the flagship and 0.2× on the lightweight model. Estimating the flagship with the lightweight multiplier can be off by 10×.
- Credit isn't withdrawable.Top-up credit is usually spend-only. Before stocking up, run one full request through on a small amount.
We don't publish relay rankings.The reason is simple: multipliers, availability and upstream sources change daily, so any static list is wrong by tomorrow. A formula you can run yourself beats a ranking that expires.