600 Tokens: A Hunyuan 4 and MiMo V2.6 Pro Receipt

01 / A 600-token trial before a bigger deposit

Two small requests can teach you more about your API spending than a long chat. Give Hunyuan 4 and MiMo V2.6 Pro the same short task, inspect both answers, and read the usage records. Your first goal is simple: learn what one useful result costs.

This walkthrough uses an illustrative receipt of 200 input tokens and 100 output tokens per model, or 600 tokens across two requests. Those are planning numbers, not measured results or a dollar quote. Actual token counts and charges depend on your requests and the current billing rates.

02 / Choose a job small enough to check yourself

Start with a task whose answer you can judge without another AI call. For example: turn a short customer message into three fields: issue, urgency, and next action. Use a fictional message about a package arriving late, and ask for a short answer with no explanation.

Send exactly the same instructions and message to both models. Keep conversation history empty. A fresh request makes the comparison easier to understand and avoids paying to resend an expanding conversation.

Judge each answer on three things: did it identify the issue, assign a reasonable urgency, and suggest an action supported by the message? A polished paragraph has little value if your application needs three usable fields.

03 / Set a $1 personal ceiling before sending requests

For this exercise, choose $1 as your personal spending ceiling. That is a suggested limit, not a claim about a minimum deposit or how many requests it buys. Check the available funding options and current rates before adding money.

CheapRelay offers pay-as-you-go access with no monthly fee or subscription. Its billing follows real token usage, including cache reads and writes. That makes the usage record worth checking even when the visible answer is short.

Sign up for CheapRelay, check the listed rates for Hunyuan 4 and MiMo V2.6 Pro, and fund only an amount you are comfortable spending. If your account offers a spending limit, use it; otherwise, track your balance and stop manually at your chosen ceiling.

04 / Make two short requests through one API key

CheapRelay provides an OpenAI-compatible API, so you can use an OpenAI SDK with your relay key and the service’s API base configured. Copy each model’s exact API identifier from the current model listing. A display name is not always the identifier accepted by an endpoint.

Keep your key in an environment variable. Set a modest output limit using the parameter supported by your chosen endpoint. An output limit restricts generation; it does not promise that the answer will use exactly that many tokens.

Send one request to each model and save the answer, usage details, and billed amount. Disable automatic retries for this first exercise if your client allows it. An unnoticed retry can turn your planned two requests into additional billable attempts.

05 / Read the receipt without inventing a price

Suppose the usage records show the following uncached totals. This is the illustrative receipt, not a report of a live test:

• ■ Hunyuan 4: 200 input tokens and 100 output tokens.
• ■ MiMo V2.6 Pro: 200 input tokens and 100 output tokens.
• ■ Combined usage: 400 input tokens and 200 output tokens, totaling 600.

If rates are quoted per million tokens, calculate each request as 0.0002 times its input rate plus 0.0001 times its output rate. Calculate the models separately, then add the charges. A shared token count does not imply a shared price.

If the receipt includes cache reads or writes, apply their listed billing treatment instead of counting them again as ordinary input. Use the recorded charge to reconcile your calculation, and check the display precision before interpreting a rounded zero as free usage.

06 / Spend the next few cents on useful answers

Choose your next trial based on cost per acceptable result. If an answer needs a second request to fix missing fields, include that request in its cost. Two requests cannot establish broad model quality, but they can reveal whether your prompt works and whether you understand the bill. Expand only after both are clear.