$0.12 for 100 light calls: a sample receipt

01 / The $0.12 receipt uses assumptions, not a price quote

A small API bill should be explainable on a small receipt. For Hunyuan 4 and MiMo V2.6 Pro, the useful starting point is not a claim about which model is cheapest. It is a count of requests, input tokens and output tokens, followed by the applicable rates.

This article uses hypothetical rates of $0.40 per million input tokens and $1.60 per million output tokens. These are arithmetic examples, not advertised prices for either model. Check the current rates before spending. No benchmark result or quality advantage is assumed here.

02 / 100 short requests produce a $0.12 example bill

Imagine a small utility that turns support notes into structured fields. Each request carries instructions and a short note, then returns a compact result. Budget 1,000 input tokens and 500 output tokens per request. Those are planning quantities, not measured token counts from a test.

• ■ Requests: 100.
• ■ Input: 100 × 1,000 = 100,000 tokens.
• ■ Output: 100 × 500 = 50,000 tokens.
• ■ Input charge: 100,000 ÷ 1,000,000 × $0.40 = $0.04.
• ■ Output charge: 50,000 ÷ 1,000,000 × $1.60 = $0.08.
• ■ Total illustrative token charge: $0.12.
• ■ Average illustrative charge per request: $0.0012.

The important line is the $0.08 output charge. Output accounts for one-third of the tokens but two-thirds of the cost under these assumptions. Counting total tokens alone would hide that difference.

03 / Give Hunyuan 4 and MiMo V2.6 Pro separate receipts

Apply the same workload to each model, but substitute each model's actual input and output rates. Equal token budgets do not establish equal prices, and identical prompts do not guarantee identical output lengths.

For either model, this workload costs 0.10 × its input rate per million tokens, plus 0.05 × its output rate per million tokens. Keep both rates in dollars. That formula is the reusable part of the receipt.

If both models happened to have the illustrative rates above, 100 calls to Hunyuan 4 would cost $0.12, and another 100 calls to MiMo V2.6 Pro would cost $0.12: $0.24 for 200 calls. That is a conditional calculation, not a statement that their prices match.

For a real comparison, send a small, representative set of tasks to each and record billed usage separately. Check whether the returned fields are usable before comparing pennies. A cheaper response that requires another paid attempt can lose its apparent advantage.

04 / A shorter response cuts this receipt to $0.08

Keep the 100 requests and 1,000 input tokens per request, but reduce output from 500 to 250 tokens. The input line stays at $0.04. Output becomes 25,000 tokens, costing $0.04 at the illustrative output rate. The new total is $0.08, a $0.04 reduction, or approximately 33.3%.

That saving comes from doing less billable work, not from an invented discount. Ask for only the fields your application needs. Avoid requesting an explanation when your program consumes only a label. Still, do not force a response so short that missing information creates retries.

05 / Check cache lines before treating $0.12 as complete

This example assumes no separately billed cache activity. CheapRelay bills per real token, including cache reads and writes, so an actual receipt should account for those categories wherever applicable. Do not silently price every input token as ordinary uncached input.

CheapRelay is pay-as-you-go, with no monthly fee or subscription. Its OpenAI-compatible routes use https://api.cheapapirelay.com/v1, which makes it practical to keep your client setup consistent while checking model-specific usage and rates.

Sign up for CheapRelay when you are ready to run a small, capped trial. Keep the first receipt boring: request count, billed token categories, applicable rates and total. For light workloads, those few lines are more useful than an unsupported promise of savings.