GPT-6 Astra relay service costs $0.5639 per million input tokens

01 / GPT-6 Astra input tokens cost $0.5639 per million

Here is the first line of the GPT-6 Astra bill: input tokens cost $0.5639 per 1,000,000 tokens on CheapRelay. The official list price for the same input line is $10.00 per million tokens. That gives a difference of $10.00 - $0.5639 = $9.4361 per million tokens. Put another way, the CheapRelay input price is $0.5639 ÷ $10.00 = 5.639% of the official list price.

That is the kind of arithmetic worth doing before you connect a project. GPT-6 Astra is not a vague bundle with an unknown multiplier attached. It has a model id, gpt-6-astra, and a price list with separate figures for input, output, and cache reads. You can see what one million input tokens will cost before you spend them.

For example, a workload processing 2,000,000 input tokens would cost 2 × $0.5639 = $1.1278 on CheapRelay. At the official input list price, the same example would cost 2 × $10.00 = $20.00. The difference is $20.00 - $1.1278 = $18.8722. This example assumes exactly two million input tokens and no other token usage.

CheapRelay sells on that published gap, not on a slogan. The site is cheapapirelay.com, and the API base is https://api.cheapapirelay.com. The model name you enter is gpt-6-astra. There is no monthly fee: you fund a prepaid balance and pay for the real tokens you use. The balance does not expire, so unused funds stay available for later work.

02 / GPT-6 Astra output tokens cost $2.82 per million

Now the second line: output tokens cost $2.82 per 1,000,000 tokens on CheapRelay. The official output list price is $50.00 per million tokens. The arithmetic is $50.00 - $2.82 = $47.18 saved per million output tokens. Divide that by the list price and the CheapRelay output rate is $2.82 ÷ $50.00 = 5.64% of list, or 94.36% below it.

Output usually gets less attention than input, even though it can dominate a bill. That is why the output line deserves its own calculation. For example, 500,000 output tokens cost 0.5 × $2.82 = $1.41 on CheapRelay. The official list price for that same example is 0.5 × $50.00 = $25.00. The difference is $25.00 - $1.41 = $23.59.

The third line is cache reads at $0.0569 per million tokens. The official cache-read list price is $1.00 per million, so the difference is $1.00 - $0.0569 = $0.9431 per million. In this example, one million cached input tokens cost $0.0569 on CheapRelay versus $1.00 at list.

Put one million tokens of each type into a single example. Total CheapRelay cost is $0.5639 + $2.82 + $0.0569 = $3.4408. The same three lines at official list prices total $10.00 + $50.00 + $1.00 = $61.00. The example saves $61.00 - $3.4408 = $57.5592. In percentage terms, $57.5592 ÷ $61.00 = 94.36%. The published gap is visible on every line, not hidden behind a subscription you have to keep renewing.

03 / What GPT-6 Astra costs against other priced models

Price comparisons are more useful when the units match. Every figure below is per 1,000,000 tokens from the current price list.

• ■ GPT-6.1 Sol charges $0.1486 for input and $0.7375 for output. Compared with GPT-6 Astra at $0.5639 and $2.82, Astra is $0.4153 higher on input and $2.0825 higher on output in this example comparison.
• ■ GPT-5.6 Sol charges $0.3056 for input, $1.71 for output, and $0.0264 for cache reads. Against Astra, the example differences are $0.2583 for input, $1.11 for output, and $0.0305 for cache reads.
• ■ Gemini 3.1 Pro lists $1.12 for input and $6.49 for output. In the example comparison, Astra is $0.5561 lower on input and $3.67 lower on output.
• ■ Claude Opus 5 lists $1.66 for input and $8.28 for output. In the example comparison, Astra is $1.0961 lower on input and $5.46 lower on output.
• ■ DeepSeek V4 Pro lists $1.01 for input, $3.03 for output, and $0.0333 for cache reads. In the example comparison, Astra is $0.4461 lower on input, $0.21 lower on output, and $0.0236 higher on cache reads.

That is not an argument for choosing one model with your eyes closed. It is an argument for checking the token mix your application actually produces. A chat assistant with long outputs may feel a different line of the bill from a classification job with short answers. CheapRelay exposes the figures so you can run the same comparison for your own traffic.

The referral program adds another line to the economics. A directly invited user contributes 20% of what they pay, and the second level contributes 6%. Those percentages are paid into a separate withdrawable rewards balance. The smallest withdrawal is ¥20, via Alipay. This is separate from the prepaid balance used for API calls, so it does not change the token prices above.

04 / Connecting GPT-6 Astra takes a base URL and a model id

GPT models use the OpenAI-compatible route under /v1 on CheapRelay. That means GPT-6 Astra works with clients such as Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL. The setup has two concrete values: use https://api.cheapapirelay.com as the API base, and enter gpt-6-astra as the model id.

That simplicity matters because a cheap token rate is not much help if the integration becomes a weekend project. You keep the client you already know, point it at the base address, and select the model by its published id. CheapRelay does not ask you to learn a new SDK for this GPT model.

The service also publishes the operating limits: 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 code gives you one clear action to take: add funds. There is no monthly fee to maintain, and the prepaid balance does not expire.

Every new account also receives $0.09 in sign-up credit, usable on any model. The credit is small, but it has a precise purpose: let you send a request and confirm the connection before committing a larger prepaid amount. It is usable on gpt-6-astra or any other model on the price list.

The refund rule is similarly concrete. If you top up and then request a refund within 12 days, the unused balance is returned, minus what you spent and the payment fee. A prepaid model with a stated refund window is easier to evaluate than a recurring charge with conditions scattered across several pages.

05 / The complete bill is visible before the first request

Let us finish with the full arithmetic. The CheapRelay price list charges $0.5639 per million input tokens, $2.82 per million output tokens, and $0.0569 per million cache reads for GPT-6 Astra. The official list prices are $10.00, $50.00, and $1.00 on those same lines.

For the example of one million tokens in each category, CheapRelay costs $3.4408 and official list pricing costs $61.00. The example difference is $57.5592, with the CheapRelay total at 5.64% of the official total. For 10,000 input tokens and 10,000 output tokens, with no cache reads, the example cost is 0.01 × $0.5639 + 0.01 × $2.82 = $0.005639 + $0.0282 = $0.033839.

Those are transparent line items. You know the model id, the endpoint, the three token rates, the official comparison, the call limits, and what happens when the balance runs out. You also know there is no monthly fee, the balance does not expire, and a new account starts with $0.09 in usable credit.

CheapRelay makes the case with numbers that can be checked: $0.5639 in, $2.82 out, $0.0569 for cache reads, and 94.36% below the official total in the one-million-token example. Then the billing stays pay-as-you-go, with prepaid funds that do not expire and a 12-day refund window for unused top-ups within 12 days. That is a relay service where the bill can be calculated before the request is sent.