$10 buys a Claude Opus 5.5 setup week on CheapRelay

01 / What $10 covers for a Claude Opus 5.5 setup week

The easiest way to judge an API is to pick a workload, count its tokens, and multiply. This article uses one clearly labelled example: a developer working with Claude Opus 5.5 on CheapRelay for five weekdays, with the same daily volume each day. The point is not to predict your own usage. The point is to show the arithmetic so you can replace the example numbers with your own.

The example developer sends 200,000 input tokens, generates 30,000 output tokens, and reads 500,000 cache-read tokens per day. That is 1,000,000 input tokens, 150,000 output tokens, and 2,500,000 cache-read tokens across the week. It is a deliberately round worksheet, not a claim about how anyone actually works.

CheapRelay prices Claude Opus 5.5 at $1.32 per 1,000,000 input tokens, $6.24 per 1,000,000 output tokens, and $0.0597 per 1,000,000 cache-read tokens. The official list prices shown for the same model are $4.00 input, $20.00 output, and $0.20 cache read, with the list column calculated at 7.2 yuan per US dollar.

• ■ Input for the example week: 1,000,000 ÷ 1,000,000 × $1.32 = $1.32
• ■ Output for the example week: 150,000 ÷ 1,000,000 × $6.24 = $0.936
• ■ Cache reads for the example week: 2,500,000 ÷ 1,000,000 × $0.0597 = $0.14925
• ■ CheapRelay example total: $1.32 + $0.936 + $0.14925 = $2.40525
• ■ Official list example total: $4.00 + $3.00 + $0.50 = $7.50
• ■ Difference: $7.50 − $2.40525 = $5.09475

Four identical example weeks cost 4 × $2.40525 = $9.621. A prepaid balance of $10 therefore covers four weeks and leaves $0.379. That is the whole pitch: show the tokens, show the unit price, show the multiplication.

02 / Connecting Claude Opus 5.5 without client detours

Before the bill matters, the request has to reach the model through a supported client. CheapRelay lists a narrow rule for Claude models, and it saves time to follow it exactly. Use the official Claude Code or Claude Desktop client, version 2.1.285 or newer. Claude Opus 5.5 is not routed through Cursor, Cherry Studio, Cline, Roo Code, OpenCode, an SDK, or cURL, and it does not run through an OpenAI-compatible endpoint.

The API base is https://api.cheapapirelay.com. When you enter the model, use the model id exactly: claude-opus-5-5. The price list shows the same id, which removes one common source of setup friction: wondering whether a display name and a request parameter are supposed to match.

CheapRelay uses prepaid balance and charges for real tokens. There is no monthly fee, and the balance does not expire. Every new account receives $0.09 in sign-up credit that can be used on any model. In the example week, that credit is $0.09 against a $2.40525 bill; it does not change the unit prices, but it does cover part of the first run.

Two operational details are worth knowing before you scale the example. Each key is limited to 120 calls per minute, and each account can run 8 calls at once. If a request returns HTTP 402, the balance is at or below zero, so top up and retry rather than rewriting the client configuration.

For people who also want a lighter model, the price list gives a useful ladder. Claude Haiku 4.5 is $0.2917 input, $1.40 output, and $0.025 cache read per 1,000,000 tokens. Claude Sonnet 5.5 is $0.8903 input, $4.45 output, and $0.0778 cache read. Claude Opus 5.5 is $1.32 input, $6.24 output, and $0.0597 cache read. The arithmetic shows where the money goes: Opus 5.5 output costs $6.24 − $4.45 = $1.79 more than Sonnet 5.5 output, while its input costs $1.32 − $0.8903 = $0.4297 more.

03 / Reading the price list line by line

The published gap is easier to trust when it is attached to a unit price. For Claude Opus 5.5 output, the official list is $20.00 per 1,000,000 tokens and the CheapRelay price is $6.24. The difference is $20.00 − $6.24 = $13.76, which is 68.8% of the list figure, matching the published statement that output is 69% under list.

For input, the official list is $4.00 and the CheapRelay price is $1.32. The difference is $2.68, or 67% of the list price. For cache reads, the list shows $0.20 and CheapRelay shows $0.0597, a difference of $0.1403. Those three comparisons are enough to describe the gap without hand-waving.

The same method works across the Claude and GPT rows, because official list prices are published for those families. Claude Sonnet 5.5 output is $4.45 against a $10.00 list, a $5.55 difference. Claude Sonnet 5 output is $4.88 against $10.00, a $5.12 difference. GPT-6.1 Sol output is $0.7375 against $10.00, a $9.2625 difference, and the price list labels it 93% under list. GPT-5.6 Sol output is $1.71 against $20.00, and GPT-6 Astra output is $2.82 against $50.00.

It is also possible to compare models without a list price by comparing each one with another listed model. Claude Opus 5 output is $8.28 per 1,000,000 tokens, so Claude Opus 5.5 output at $6.24 is $2.04 lower. Gemini 3.1 Pro output is $6.49, which is $0.25 above Claude Opus 5.5 output. Kimi K3 output is $4.67, which is $1.57 below Claude Opus 5.5 output. These are direct comparisons between listed prices, not estimates.

Gemini follows a different client path. Gemini models use the native protocol under /v1beta through Gemini CLI, not an OpenAI-compatible endpoint. GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax, and the other listed models can use an OpenAI-compatible client under /v1, including Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL. Keeping those paths straight avoids a wasted afternoon.

04 / What the prepaid balance does with unused money

A prepaid balance changes the shape of the decision. There is no monthly fee, and money already in the account does not expire. If the example week ends with $7.59475 remaining from a $10 top-up, that amount stays on the balance for later token usage. You are not paying for a fixed period of access; you are paying for the requests you actually send.

The refund rule is straightforward: unused balance can be refunded within 12 days of the top-up, minus what was spent and minus the payment fee. For the example, a $10 top-up with $2.40525 spent would leave $7.59475 unused before the payment fee is deducted. Keeping that 12-day window in view is useful if you are still testing a setup rather than committing to a steady workload.

CheapRelay also publishes a referral structure: 20% of what a directly invited user pays, plus 6% from the second level, paid into a withdrawable rewards balance. The smallest withdrawal is ¥20, via Alipay. In a worked example, if a directly invited user pays $10, the referral earns 20% × $10 = $2 into the rewards balance. If someone below that invitee pays $10, the second-level share is 6% × $10 = $0.60. Those examples do not change the API price; they describe a separate rewards account.

05 / A practical first-day checklist for the example workload

Start with the client rule, because it removes the most common setup detour. Install or update the official Claude Code or Claude Desktop client to version 2.1.285 or newer. Set the CheapRelay API base to https://api.cheapapirelay.com, and enter claude-opus-5-5 as the model id. Do not route Claude Opus 5.5 through an OpenAI-compatible client.

Next, keep a token counter beside the test. For this example, the daily target is 200,000 input, 30,000 output, and 500,000 cache-read tokens. The daily CheapRelay cost is 0.2 × $1.32 + 0.03 × $6.24 + 0.5 × $0.0597 = $0.264 + $0.1872 + $0.02985 = $0.48105. Five days at $0.48105 gives $2.40525, matching the weekly total shown earlier.

Then check the balance before you scale up. The $0.09 sign-up credit can be spent on any model, the account has no monthly fee, and the prepaid balance does not expire. Stay within 120 calls per minute per key and 8 concurrent calls per account. If HTTP 402 appears, the balance is at or below zero; add funds and resend the request.

Finally, put your own numbers beside the example numbers. If your week uses 500,000 input tokens and 80,000 output tokens with the same 2,500,000 cache-read tokens, then 0.5 × $1.32 + 0.08 × $6.24 + 2.5 × $0.0597 = $0.66 + $0.4992 + $0.14925 = $1.30845. The method is the same each time: multiply token counts by the published unit price, add the lines, and compare the result with the official list price where one is shown. CheapRelay makes that arithmetic visible before you commit a larger budget.