Check the relay service bill for Qwen3.8 Max on CheapRelay

01 / The Qwen3.8 Max price list and the bill it creates

I start every API purchase with arithmetic, not adjectives. On CheapRelay, Qwen3.8 Max uses the model id qwen3.8-max, and the price list charges $0.5722 per 1,000,000 input tokens, $1.72 per 1,000,000 output tokens, and $0.0681 per 1,000,000 cache-read tokens. There is no monthly fee. You fund a prepaid balance, and the system bills the real tokens your request uses.

Those three numbers are enough to build a bill before you spend anything. Suppose an example request uses 100,000 input tokens and 200,000 output tokens. The input cost is 100,000 divided by 1,000,000, multiplied by $0.5722, which equals $0.05722. The output cost is 200,000 divided by 1,000,000, multiplied by $1.72, which equals $0.344. Add them and the example bill is $0.40122.

That is the useful starting point: the price per million multiplied by the fraction of a million tokens you actually use. CheapRelay does not ask you to guess a package size or commit to a fixed monthly allowance. The balance does not expire, so unused money remains available for later work.

• ■ Qwen3.8 Max input: $0.5722 per 1,000,000 tokens
• ■ Qwen3.8 Max output: $1.72 per 1,000,000 tokens
• ■ Qwen3.8 Max cache read: $0.0681 per 1,000,000 tokens
• ■ Model id to enter: qwen3.8-max

When I evaluate a relay service, I want the unit prices visible before I configure a client. CheapRelay publishes those figures directly on its price list, so the first-day budget can be calculated from the same numbers that will later appear on the bill.

02 / A before-and-after bill when cache reads enter the picture

Here is the “before” version of the example bill: 100,000 input tokens plus 200,000 output tokens, with no cache-read line. As calculated above, that comes to $0.05722 plus $0.344, or $0.40122.

Now build the “after” version for an example run where 1,000,000 tokens are read from cache. The cache-read line is 1,000,000 divided by 1,000,000, multiplied by $0.0681, which equals $0.0681. The full example bill becomes $0.05722 plus $0.344 plus $0.0681, or $0.46932.

The point is not that every workload has the same mix. The point is that each line has a visible multiplier. If your example workload instead reads 10,000,000 cache tokens, that line is 10 multiplied by $0.0681, or $0.681. You can scale the arithmetic before switching traffic over.

For another model comparison using the same 100,000 input and 200,000 output example, Qwen3.7 Plus costs $0.01208 for the input and $0.07250 for the output, totaling $0.08458. Qwen3.8 Max costs $0.40122 for that same token mix. The difference is $0.40122 minus $0.08458, or $0.31664. Qwen3.8 Max is the more expensive of those two figures, and the bill shows exactly where that difference comes from: its higher input and output rates.

Compare Qwen3.8 Max with Qwen3.6 Plus Uncensored on the same example. Qwen3.6 Plus Uncensored costs $0.03431 for 100,000 input tokens and $0.366 for 200,000 output tokens, totaling $0.40031. Qwen3.8 Max totals $0.40122, so the difference is $0.00091. On that example, the two bills are close, but the comparison does not apply to every mix because Qwen3.6 Plus Uncensored has no cache-read rate on the price list, while Qwen3.8 Max does.

03 / Connecting Qwen3.8 Max through the right client

Qwen belongs to the OpenAI-compatible group on CheapRelay. That means you can use Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL. These are the clients to pair with Qwen3.8 Max, along with GPT, DeepSeek, Kimi, GLM, Grok, MiniMax, and the other models listed for OpenAI-compatible use.

The API base is https://api.cheapapirelay.com, and the OpenAI-compatible route is under /v1. The model name to enter is qwen3.8-max. Keep those three details separate in your setup: base address, route, and model id. Mixing them is usually a five-minute typo rather than a pricing problem.

One warning saves a longer debugging session: Claude models on CheapRelay run only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. They are not connected through Cursor, Cherry Studio, Cline, Roo Code, OpenCode, an SDK, cURL, or an OpenAI-compatible route. Gemini models use their native protocol under /v1beta through Gemini CLI. For this article, though, the simple path is Qwen3.8 Max through an OpenAI-compatible client under /v1.

After a request succeeds, check the token counts reported by the client and compare them with your own example estimate. If the example expected 100,000 input and 200,000 output tokens, then $0.40122 is the arithmetic to compare against. If the request also read cached tokens, add the cache line at $0.0681 per million. This makes a bill review quick: split it into input, output, and cache, then multiply each count by its published rate.

04 / Funding the balance, using the credit, and keeping the cash safe

CheapRelay uses a prepaid balance. There is no monthly fee, and the balance does not expire. Every new account receives $0.09 in sign-up credit, usable on any model. That credit is enough to run a small example request and confirm that your base URL, route, and model id are correct before committing a larger top-up.

The refund rule is plain: unused balance within 12 days of the top-up can be refunded, minus what was spent and the payment fee. I treat that window as a testing period. Put in a small amount, verify the bill, and then decide whether to add more. If you do not use the money, the unused portion inside the stated window is not trapped there.

Referrals also have visible arithmetic. A directly invited user contributes 20% of what they pay, while the second level contributes 6%. Those percentages land in a withdrawable rewards balance. The smallest withdrawal is ¥20, and withdrawals go through Alipay. This is separate from the prepaid balance used for API calls, so I keep the two pots distinct when checking my accounts.

CheapRelay also makes rate behavior visible: 120 calls per minute per key, and 8 calls at once per account. If you receive HTTP 402, the balance is at or below zero. That status gives you one specific action—add funds—rather than a vague failure to troubleshoot.

05 / Putting the Qwen3.8 Max numbers beside the rest of the price list

CheapRelay’s price list is easier to judge when each figure is read beside another real figure. Qwen3.8 Max input is $0.5722 per million, while Qwen3.7 Plus input is $0.1208. Qwen3.8 Max output is $1.72 per million, while Qwen3.7 Plus output is $0.3625. So Qwen3.8 Max carries the higher rate on both sides of that comparison. Whether that is worth paying depends on the token mix and the quality of result your task needs.

For models that have an official list price, the arithmetic can be even clearer. GPT-6.1 Sol lists officially at $2.00 input and $10.00 output per million. On CheapRelay, GPT-6.1 Sol is $0.1486 input and $0.7375 output, with a cache read of $0.0347 against an official $0.10. For an example 100,000 input and 200,000 output workload, the official figure is $0.20 plus $2.00, or $2.20. The CheapRelay figure is $0.01486 plus $0.14750, or $0.16236. The example difference is $2.20 minus $0.16236, or $2.03764.

Claude Sonnet 5.5 shows a similar published gap: official list is $2.00 input and $10.00 output per million, while CheapRelay charges $0.8903 and $4.45. On the same example workload, official pricing produces $2.20 and CheapRelay produces $0.08903 plus $0.89, or $0.97903. The example difference is $1.22097.

For Qwen3.8 Max itself, there is no official list price in this price list, so I do not manufacture one. I compare it with the Qwen, GPT, DeepSeek, Kimi, GLM, Grok, MiniMax, Gemini, and Claude figures that are actually listed. Qwen3.8 Max output at $1.72 is below GPT-6.1 Sol at $0.7375? No: $1.72 is higher than $0.7375. It is also below GPT-5.6 Sol at $1.71 by only $0.01 per million. Those comparisons are enough to show where it sits without pretending a reference price that is not published.

My first-day checklist is therefore short. Confirm that Qwen3.8 Max is $0.5722 input, $1.72 output, and $0.0681 cache read per million. Enter qwen3.8-max through an OpenAI-compatible client under /v1. Use the $0.09 credit for a small test, compare the resulting token counts with your arithmetic, and keep the prepaid balance only as long as your plan needs it. CheapRelay makes the purchase legible because the unit prices are published, the bill follows token counts, and there is no monthly fee between you and the final number.