Claude Sonnet 5 pricing: $0.9764 input on CheapRelay

01 / Claude Sonnet 5 pricing on CheapRelay, one line at a time

Model prices look complicated until you put each number on its own line. Here is the Claude Sonnet 5 row from the CheapRelay price list, with the model id you enter set to claude-sonnet-5.

• ■ Input: $0.9764 per 1,000,000 tokens.
• ■ Output: $4.88 per 1,000,000 tokens.
• ■ Cache read: $0.0861 per 1,000,000 tokens.

That is the whole pricing structure. There is no monthly fee, no seat charge, and no package to buy. CheapRelay uses a prepaid balance, and you pay according to the real tokens your request consumes. Your balance does not expire, so unused money stays available for later work.

The arithmetic is straightforward. If a request processes 100,000 input tokens, the input charge is $0.9764 divided by 10, which equals $0.09764. If it generates 20,000 output tokens, the output charge is $4.88 multiplied by 0.02, which equals $0.0976. Add them, and the request costs $0.19524 before any cache read. No estimation, no hidden subscription, no guessing which tier you landed on.

02 / What the official Claude list price changes

The price list also shows official Claude list prices, converted at 7.2 yuan per US dollar. For Claude Sonnet 5, those figures are $2.00 for input, $10.00 for output, and $0.20 for cache reads, each per 1,000,000 tokens.

Now do the subtraction instead of accepting a vague discount claim.

• ■ Input gap: $2.00 minus $0.9764 equals $1.0236. Divide that by $2.00, and the CheapRelay input price is 51.18% below the listed figure.
• ■ Output gap: $10.00 minus $4.88 equals $5.12. Divide that by $10.00, and the output price is 51.2% below the listed figure.
• ■ Cache-read gap: $0.20 minus $0.0861 equals $0.1139. Divide that by $0.20, and the cache-read price is 56.95% below the listed figure.

Those are direct comparisons with the published Claude figures on the price list. CheapRelay is not presented as an official Claude provider or partner; it is a relay with its own published pricing and a measurable gap to the official list price.

03 / Two example bills, with every step visible

Example one: suppose a typical coding session consumes 1,000,000 input tokens, 200,000 output tokens, and 100,000 cache-read tokens.

• ■ Input: 1,000,000 divided by 1,000,000 times $0.9764 equals $0.9764.
• ■ Output: 200,000 divided by 1,000,000 times $4.88 equals $0.976.
• ■ Cache read: 100,000 divided by 1,000,000 times $0.0861 equals $0.00861.
• ■ Total on CheapRelay: $0.9764 plus $0.976 plus $0.00861 equals $1.96101.

At the official list figures, the same example would be $2.00 plus $2.00 plus $0.02, which equals $4.02. The difference is $4.02 minus $1.96101, or $2.05899 for that one example workload.

Example two: a lighter session with 500,000 input tokens and 100,000 output tokens costs $0.4882 plus $0.488, which equals $0.9762 on CheapRelay. The official list figures would produce $1.00 plus $1.00, or $2.00. The example saves $1.0238.

Example three: a heavier month of usage with 10,000,000 input tokens and 2,000,000 output tokens costs $9.764 plus $9.76, which equals $19.524 on CheapRelay. The same example at the official list figures would be $20 plus $20, or $40. The example gap is $20.476.

These examples are illustrations, not guarantees about your bill. Your real cost follows your real token counts, which is exactly why per-token pricing is easier to audit than a subscription.

04 / Paying for it without a monthly commitment

CheapRelay starts with a simple funding model: add money to a prepaid balance, then draw against it as tokens are processed. There is no monthly fee. The balance does not expire, so you do not lose unused credit because a quiet month passes.

Every new account receives $0.09 in sign-up credit, usable on any model. That is enough to run a small test and confirm that the setup works before committing more money. It is a practical first step rather than a large headline number.

There are also two clear billing signals. An HTTP 402 response means the balance is at or above the zero line, so you know exactly when a top-up is needed. Per key, the limit is 120 calls per minute, and an account can run 8 calls at once. Those are published operating limits, not extra charges.

The refund policy is equally concrete. If you top up and then request a refund within 12 days, the unused balance can be refunded, minus what you already spent and the payment fee. That keeps the process tied to actual usage instead of vague terms.

Referrals also have visible arithmetic. A directly invited user generates a 20% rewards share based on what they pay, while the second level contributes 6%. Those rewards enter a withdrawable balance, and the smallest withdrawal is ¥20 through Alipay.

05 / How to configure Claude Sonnet 5 correctly

Getting a good price is only half the job; the request has to reach the right client. Claude models on CheapRelay work with the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. That is the supported pairing for claude-sonnet-5.

Do not expect Claude Sonnet 5 to run through Cursor, Cherry Studio, Cline, Roo Code, OpenCode, an SDK, or cURL. Those are not the supported route for Claude models, and an OpenAI-compatible setup is not the answer here either.

The workflow is short. Create an account, use the $0.09 credit if you want a quick test, add funds when you are ready, and configure the supported Claude client with the CheapRelay endpoint at api.cheapapirelay.com. The API base is https://api.cheapapirelay.com. Enter claude-sonnet-5 as the model id, then check the token usage shown by the client against the three rates above.

For comparison, if you later work with GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax, or another OpenAI-compatible model, those models use a different supported route: an OpenAI-compatible client under /v1, such as Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL. Gemini is separate again, using the native protocol under /v1beta through Gemini CLI. Matching the client to the model avoids wasted debugging time.

06 / Where Claude Sonnet 5 sits among the other Claude prices

The price list lets you compare Claude models directly. Claude Sonnet 5.5, with model id claude-sonnet-5-5, lists input at $0.8903, output at $4.45, and cache read at $0.0778 per 1,000,000 tokens. Claude Sonnet 5 lists $0.9764, $4.88, and $0.0861 respectively. Since each Sonnet 5 figure is higher than its Sonnet 5.5 counterpart, Sonnet 5.5 costs less on all three lines in the price list.

Claude Opus 5, with model id claude-opus-5, lists $1.66 input, $8.28 output, and $0.1222 cache read. Claude Sonnet 5 lists $0.9764, $4.88, and $0.0861. Each Sonnet 5 figure is lower than the corresponding Opus 5 figure.

Apply that to the earlier example of 1,000,000 input tokens, 200,000 output tokens, and 100,000 cache-read tokens. Claude Opus 5 would cost $1.66 plus $1.656 plus $0.01222, which equals $3.32822. Claude Sonnet 5 costs $1.96101 for the same example. Claude Sonnet 5.5 costs $0.8903 plus $0.89 plus $0.00778, which equals $1.78808.

That gives you a clean choice. If you want the lower published Claude rates in this comparison, Claude Sonnet 5.5 fits that example at $1.78808. If you specifically need the claude-sonnet-5 model id, its $1.96101 example total still sits below the $3.32822 example total for Claude Opus 5 and well below the $4.02 official-list example for Claude Sonnet 5.

CheapRelay makes the decision easy to check: open the price list, pick the model, apply the token counts, and read the total. For claude-sonnet-5, the arithmetic is $0.9764 for input, $4.88 for output, and $0.0861 for cache reads per 1,000,000 tokens, funded from a prepaid balance with no monthly fee.