Grok 4.7 via CheapRelay API relay: $1.52 per 1M output tokens

01 / The price list entry for Grok 4.7

I price API work the way I price electricity: find the unit price, count the units, multiply, and check the receipt. On the CheapRelay price list today, Grok 4.7 has three separate rates per 1,000,000 tokens. Input costs $0.5056, output costs $1.52, and a cache read costs $0.1264.

That is the whole starting point. There is no monthly fee, no seat charge, and no package to buy. CheapRelay uses a prepaid balance and charges per real token. The model id to enter is grok-4.7, the API base is https://api.cheapapirelay.com, and Grok works through an OpenAI-compatible client under /v1. That means you can use the setup your development habits already have, such as Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL.

The arithmetic below uses a worked week. It is an example, not a usage claim or a forecast. The point is to show how three line items become a bill you can inspect.

02 / Example week: the traffic we will price

Here is the example week. On Monday, the job sends 120,000 input tokens, receives 25,000 output tokens, and reads 300,000 cached tokens. Tuesday uses 90,000 input, 18,000 output, and 250,000 cache-read tokens. Wednesday uses 150,000 input, 30,000 output, and 400,000 cache-read tokens. Thursday uses 80,000 input, 12,000 output, and 180,000 cache-read tokens. Friday is the busy day at 200,000 input, 40,000 output, and 600,000 cache-read tokens. Saturday uses 60,000 input, 10,000 output, and 150,000 cache-read tokens. Sunday uses 100,000 input, 20,000 output, and 320,000 cache-read tokens.

• ■ Total input for the example: 800,000 tokens.
• ■ Total output for the example: 155,000 tokens.
• ■ Total cache reads for the example: 2,200,000 tokens.
• ■ Total calls for the example: 7 daily batches, within the limits of 120 calls per minute per key and 8 simultaneous calls per account.

Every number in those totals comes from adding the seven example days. No rounding fudge, no hidden subscription, no separate platform fee. The next section prices each line.

03 / Example week: Grok 4.7 priced line by line

Input is the first line. The example uses 800,000 input tokens, which is 0.8 of one million. At $0.5056 per million, the calculation is 0.8 × $0.5056 = $0.40448.

Output is the second line. The example uses 155,000 output tokens, or 0.155 of one million. At $1.52 per million, that is 0.155 × $1.52 = $0.23560.

Cache reads are the third line. The example uses 2,200,000 cached tokens, or 2.2 of one million. At $0.1264 per million, that is 2.2 × $0.1264 = $0.27808.

Now add the three lines: $0.40448 + $0.23560 + $0.27808 = $0.91816. That is the total for this example week with Grok 4.7 on CheapRelay. It is less than one dollar for 800,000 input tokens, 155,000 output tokens, and 2,200,000 cache reads.

For a daily view, the same arithmetic gives $0.136592 on Monday, $0.104464 on Tuesday, $0.17200 on Wednesday, $0.081440 on Thursday, $0.237760 on Friday, $0.064496 on Saturday, and $0.121408 on Sunday. Those seven daily figures add up to $0.91816. A developer can therefore check the bill by hand, line by line, without a spreadsheet full of mystery adjustments.

04 / Example week: the same traffic on Grok 4.6

Before choosing a model, I compare the actual figures. Grok 4.6 lists input at $0.4653, output at $1.39, and cache read at $0.1167 per million tokens. Applied to the same example week, the arithmetic is:

• ■ Input: 0.8 × $0.4653 = $0.37224.
• ■ Output: 0.155 × $1.39 = $0.21545.
• ■ Cache read: 2.2 × $0.1167 = $0.25674.
• ■ Example total: $0.37224 + $0.21545 + $0.25674 = $0.84443.

Grok 4.7 therefore costs $0.07373 more for this example week, or about 8.7% more than Grok 4.6. That difference comes from three published unit prices: $0.5056 versus $0.4653 for input, $1.52 versus $1.39 for output, and $0.1264 versus $0.1167 for cache reads.

Now compare Grok 4.7 with Qwen3.8 Max, whose listed rates are $0.5722 input, $1.72 output, and $0.0681 cache read. On the example week, Qwen3.8 Max costs 0.8 × $0.5722 + 0.155 × $1.72 + 2.2 × $0.0681 = $0.45776 + $0.26660 + $0.14982 = $0.87418. Grok 4.7 costs $0.04398 more for this workload, even though its cache-read rate is lower. That is the value of doing the arithmetic instead of repeating slogans: the right model depends on the token mix you actually generate.

05 / How the published price gap shows up on the same example

The price list also publishes official list prices for some Claude and GPT models, so those gaps can be calculated directly. Take GPT-6.1 Sol as a reference. Its official list prices are $2.00 input, $10.00 output, and $0.50 cache read per million. On the same example week, the official list cost would be 0.8 × $2.00 + 0.155 × $10.00 + 2.2 × $0.50 = $1.60 + $1.55 + $1.10 = $4.25.

The GPT-6.1 Sol rates on CheapRelay are $0.1486 input, $0.7375 output, and $0.0347 cache read. The example cost there is 0.8 × $0.1486 + 0.155 × $0.7375 + 2.2 × $0.0347 = $0.11888 + $0.1143125 + $0.07634 = $0.3095325. That is the kind of published gap CheapRelay wants customers to see and verify: $4.25 at official list versus $0.3095325 through CheapRelay for the same token counts.

Grok 4.7 has no official list price shown on this price list, so there is no official Grok figure to subtract from its $0.5056, $1.52, and $0.1264 rates. The honest comparison for Grok is the arithmetic above: $0.91816 for the example week on Grok 4.7, against $0.84443 on Grok 4.6 and $0.87418 on Qwen3.8 Max.

06 / Starting on CheapRelay without a monthly fee

Setup is deliberately plain. Create an account, receive the $0.09 sign-up credit, add funds to the prepaid balance, and send requests to the API base with the model id grok-4.7. The credit works on any model, so it can go against this example week’s Grok 4.7 bill. If the entire $0.09 is applied, the remaining example cost is $0.91816 − $0.09 = $0.82816.

The balance does not expire, and CheapRelay has no monthly fee. If a request returns HTTP 402, the balance is at or below zero, so the fix is visible: add funds. The prepaid model also has a clear refund rule. Unused balance can be refunded within 12 days of a top-up, minus what has already been spent and the payment fee.

CheapRelay also supports referrals. A directly invited user contributes 20% of what they pay, and the second level contributes 6%. Those amounts go into a withdrawable rewards balance, with the smallest withdrawal at ¥20 through Alipay. That is separate from the prepaid spending balance, so you can see both the money used for API work and the reward money separately.

The final check is the receipt. In this example week, Grok 4.7 cost $0.91816: $0.40448 for input, $0.23560 for output, and $0.27808 for cache reads. The rates came straight from the price list, the usage came from the example, and every line multiplied out. That is the CheapRelay proposition in arithmetic form: know the unit price, use a prepaid balance, connect through an OpenAI-compatible client, and verify the total yourself.