Cherry Studio, One Key: A Token-by-Token Receipt

01 / One Connection, Two Models, Separate Line Items

Putting several models behind one key should simplify configuration, not obscure spending. The useful question is not how many models appear in your menu. It is how much each completed task costs, including input, output, cache activity, and any additional attempts.

CheapRelay offers a pay-as-you-go connection to multiple model families without a monthly fee or subscription. In Cherry Studio, that means you can configure a relay provider and select models through that connection rather than maintain a separate credential for every upstream service. Cherry Studio supports custom compatible providers and model selection within their configuration. ([cherryai.com](https://www.cherryai.com/docs/en/pre-basic/providers/?utm_source=openai))

For a concrete example, consider a workspace with Gemini 2.5 Flash and Claude Haiku 4.5. These are two available choices, not a claim that either wins a particular benchmark. Keep their usage on separate receipt lines even though they share one relay key.

02 / Connect the Relay Before Sending a Large Prompt

Open Settings, then Model Provider. Add a custom OpenAI-compatible provider, name it CheapRelay, and enter your relay key. Cherry Studio documents custom providers for compatible endpoints, so you do not need to pretend the relay is a direct upstream account. ([cherryai.com](https://cherryai.com/docs/en/pre-basic/providers/zi-ding-yi-fu-wu-shang/?utm_source=openai))

Use the API host supplied by CheapRelay. Its OpenAI-compatible routes sit under /v1. Watch how your Cherry Studio version handles that suffix: the desktop documentation says standard configurations append /v1/chat/completions automatically. Avoid accidentally producing a doubled version path. ([cherryai.com](https://cherryai.com/docs/en/pre-basic/settings/providers/?utm_source=openai))

Sync the available models, add the ones you intend to use, enable the provider, and send a short test message. Cherry Studio documents this sequence and offers a model connection check. A successful short response verifies the basic connection before you send a lengthy document. ([cherryai.com](https://www.cherryai.com/docs/en/pre-basic/providers/?utm_source=openai))

03 / An Illustrative Receipt Totals $1.32

Live CheapRelay prices were not independently verified for this article. The following rates are deliberately hypothetical budgeting inputs, not advertised prices, discounts, or a real customer invoice. Replace them with the current rates for each selected model before making a purchase decision.

Assume one batch uses these example rates per million tokens: $0.50 for uncached input, $2.00 for output, $0.10 for cache reads, and $0.75 for cache writes. For this worksheet, the four token buckets are separate; no token is counted twice.

• ■ Uncached input: 1,000,000 tokens × $0.50 per million = $0.50.
• ■ Output: 300,000 tokens × $2.00 per million = $0.60.
• ■ Cache reads: 1,500,000 tokens × $0.10 per million = $0.15.
• ■ Cache writes: 100,000 tokens × $0.75 per million = $0.075.
• ■ Illustrative usage subtotal: $1.325, or about $1.33.
• ■ At 100 completed tasks: $0.01325 per task, approximately 1.33 cents.

Notice the receipt correction: the exact subtotal is $1.325, not $1.32 under ordinary cent rounding. Keep full precision while calculating and round only for display. Tiny rounding differences become noticeable when you reconcile many small requests.

04 / Two Models Require Two Sets of Rates

The sample above describes one rate schedule. A genuine multi-model receipt needs one calculation per model, using that model's applicable rates. Add the subtotals afterward; do not average prices first unless you also weight them by actual token usage.

Using entirely illustrative subtotals, a $1.325 batch plus a $2.40 batch costs $3.725, displayed as $3.73. Across 150 completed tasks, that is approximately 2.48 cents per task. Neither subtotal is a price claim about the two named models.

CheapRelay bills real token usage, including cache reads and writes. Therefore, a useful reconciliation records the model identifier, token categories, applicable rates, and charge. Do not assume cached input is free or that every repeated prompt qualifies for caching.

05 / Make the Next Receipt Smaller, Not Less Visible

Set an output-length target, remove irrelevant pasted material, and count extra attempts in the same task total. In the hypothetical schedule, cutting output from 300,000 to 150,000 tokens reduces that line from $0.60 to $0.30. That is arithmetic, not a promised quality-preserving saving.

Sign up for CheapRelay, connect one key in Cherry Studio, and run a small measured batch before expanding usage. The practical benefit is consolidated access; the financial discipline is keeping every model's receipt readable.