$0.3056: GPT-5.6 Sol on a CheapRelay relay service

01 / The before bill: GPT-5.6 Sol at official list price

Here is the number that makes a relay service worth checking: GPT-5.6 Sol costs $0.3056 per 1,000,000 input tokens on CheapRelay, $1.71 per 1,000,000 output tokens, and $0.0264 per 1,000,000 cache-read tokens. The official list prices for the same token categories are $4.00, $20.00, and $0.40. The published comparison uses 7.2 yuan per US dollar.

Let us make the gap visible without hand-waving. On output tokens, $20.00 minus $1.71 is $18.29. Divide that saving by $20.00 and you get 91.45%, which rounds to the published 91% under list. On input tokens, $4.00 minus $0.3056 is $3.6944, or 92.36% below the listed figure. On cache reads, $0.40 minus $0.0264 is $0.3736, or 93.4% below list.

Now build an example bill. Suppose a workload sends 1,000,000 input tokens, receives 200,000 output tokens, and reads 500,000 cached tokens. At the official list price, the example input cost is $4.00, the output cost is $4.00, and the cache-read cost is $0.20. The example official-list bill is therefore $8.20.

That is the “before” figure. It is not a subscription and it does not include a monthly fee. It is simply the cost of the same token volume priced at the published GPT-5.6 Sol list rates.

02 / The after bill: the same workload through CheapRelay

Run the identical example through CheapRelay. The 1,000,000 input tokens cost $0.3056. The 200,000 output tokens cost one-fifth of $1.71, which is $0.342. The 500,000 cache-read tokens cost half of $0.0264, which is $0.0132. Add them: $0.3056 plus $0.342 plus $0.0132 equals $0.6608.

So the example bill falls from $8.20 to $0.6608. The difference is $7.5392 for that workload. That is arithmetic, not a vague promise about savings.

Here is another example that keeps the arithmetic easy to audit. A request with 100,000 input tokens and 10,000 output tokens costs $0.03056 plus $0.0171, or $0.04766 through CheapRelay. Priced at the official list rates, the same example would be $0.40 plus $0.20, or $0.60. CheapRelay handles the billing as pay-per-real-token usage against a prepaid balance.

There is no monthly fee, and the balance does not expire. Every new account also receives $0.09 in signup credit, usable on any model. That credit is small, but it gives you something concrete to test: create the account, add the balance, and watch the usage meter reduce it as real tokens pass through the API.

For refunds, unused balance can be reclaimed within 12 days of the top-up, with the amount spent and the payment fee deducted. That policy makes the first-day arithmetic easier to reason about because the balance is prepaid rather than tied to a recurring charge.

03 / First-day setup: base URL, model id, and the right client

GPT-5.6 Sol is an OpenAI-compatible model, so point your client at the CheapRelay API base, https://api.cheapapirelay.com, and use the /v1 path. When the interface asks for a model, enter the model id exactly: gpt-5.6-sol.

That pairing gives you room to use the tooling you already have. Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, and cURL are all compatible choices for GPT models, including GPT-5.6 Sol. Codex can also be used with GPT through the OpenAI-compatible route.

Two operational limits are worth putting on the screen before your first test: each key is limited to 120 calls per minute, and each account can run 8 calls at once. If you receive HTTP 402, the account balance is at or below zero. That is a straightforward signal to top up rather than troubleshoot blindly.

Model routing matters, so keep the client rules separate. Gemini models use the native protocol under /v1beta through Gemini CLI. Claude models use the official Claude Code or Claude Desktop client, version 2.1.285 or newer. GPT-5.6 Sol does not need those native routes; it belongs with the OpenAI-compatible /v1 setup described above.

04 / Where the output price sits on the CheapRelay price list

GPT-5.6 Sol is not priced in isolation, so let us compare the relevant figures before choosing it. Its output rate is $1.71 per 1,000,000 tokens. GPT-6.1 Sol is $0.7375, so GPT-6.1 Sol has a lower output rate by $1.71 minus $0.7375, or $0.9725 per 1,000,000 tokens. If your workload is output-heavy and GPT-6.1 Sol meets the requirement, that difference is substantial.

On input, GPT-5.6 Sol is $0.3056, while GPT-6.1 Sol is $0.1486. The difference is $0.1570 per 1,000,000 input tokens. Both models have published official list prices: GPT-5.6 Sol lists at $4.00 input and $20.00 output, while GPT-6.1 Sol lists at $2.00 input and $10.00 output. The key point is that the comparison is always between the two rows on the price list, not between a listed model and a borrowed price from somewhere else.

Other GPT rows give you more context. GPT-6 Astra is $0.5639 input, $2.82 output, and $0.0569 cache read. Its official list prices are $10.00, $50.00, and $1.00. On output, $50.00 minus $2.82 is $47.18, which is 94.36% below list and rounds to the published 94%. GPT-5.6 Sol remains the more economical GPT-5.6 Sol choice at $1.71 output, but the arithmetic gives you a clean way to select according to the job.

Non-GPT models have their own price-list rates, but no official list prices are shown for them here, so the honest comparison is model to model. DeepSeek V4 Flash is $0.2389 input and $0.9556 output. Gemini 2.5 Flash is $0.1778 input and $1.42 output. Qwen3.7 Plus is $0.1208 input and $0.3625 output. You can put your own token counts into those figures without pretending they carry an official-list comparison that is not published.

05 / Making the account pay for itself

The first-day checklist is short. Create a CheapRelay account and claim the $0.09 signup credit. Add a prepaid balance. Configure the API base as https://api.cheapapirelay.com and use /v1 for GPT-5.6 Sol. Enter gpt-5.6-sol as the model id, choose an OpenAI-compatible client, and send a small test request while watching the balance move.

Referrals are also priced clearly. A directly invited user contributes 20% of what they pay, and the second level contributes 6%. Those percentages go into a withdrawable rewards balance, with a ¥20 minimum withdrawal through Alipay. There is no monthly fee attached to the account, and unused prepaid balance does not expire.

For the worked example, keep the bill in front of you: $8.20 at the official list rate becomes $0.6608 through CheapRelay for 1,000,000 input tokens, 200,000 output tokens, and 500,000 cache-read tokens. Then adjust the token counts to your own work. The model id is simple, the client path is standard, and the bill is built from real token prices rather than a subscription label.