Claude Sonnet 5.5 setup: check the $0.8903 input price before paying

01 / The Claude Sonnet 5.5 entry on the CheapRelay price list

Claude Sonnet 5.5 arrives on CheapRelay with one of the shorter bills in the catalog, but the number that matters is the token rate, not the headline. On the price list, the model id is claude-sonnet-5-5, and the rates per 1,000,000 tokens are $0.8903 for input, $4.45 for output, and $0.0778 for a cache read. There is no monthly fee: you top up a prepaid balance and pay per real token as you use it.

Here is the arithmetic on one ordinary request. Suppose a call carries 10,000 input tokens and returns 2,000 output tokens. Input is 10,000 ÷ 1,000,000 = 0.01 of a million, so it costs 0.01 × $0.8903 = $0.008903. Output is 2,000 ÷ 1,000,000 = 0.002, so it costs 0.002 × $4.45 = $0.0089. The total is $0.017803, roughly $0.0178 for 12,000 tokens of real work. Add a cache read of 100,000 tokens and 0.1 × $0.0778 = $0.00778. Small numbers, but they compound across every request in a day.

CheapRelay also credits $0.09 to every new account at sign-up, usable on any model, and the balance does not expire. A developer can spend that credit on Claude Sonnet 5.5 at the same rates, with no subscription attached.

02 / How the Claude Sonnet 5.5 rate compares with the official list price

The price list gives an official list price for Claude Sonnet 5.5: $2.00 input, $10.00 output, and $0.10 cache read per 1,000,000 tokens, converted at 7.2 yuan per US dollar. CheapRelay charges $0.8903 input, $4.45 output, and $0.0778 cache read. Do the subtraction:

• ■ Input: $2.00 − $0.8903 = $1.1097 saved, so the relay rate is about 56% below list.
• ■ Output: $10.00 − $4.45 = $5.55 saved, also about 56% below list.
• ■ Cache read: $0.10 − $0.0778 = $0.0222 saved, about 22% below list.

Converted at 7.2 yuan per dollar, the official list price becomes ¥14.40 input, ¥72.00 output, and ¥0.72 cache read per million tokens. The CheapRelay rates become 0.8903 × 7.2 = ¥6.41016 input, 4.45 × 7.2 = ¥32.04 output, and 0.0778 × 7.2 = ¥0.56016 cache read. The gaps are ¥7.98984 on input, ¥39.96 on output, and ¥0.15984 on cache reads, for every million tokens.

Output usually dominates an agent bill, because a coding assistant reads a lot and writes a lot. That is why the ¥39.96 output gap is the line worth watching.

03 / A worked example: one million tokens each way

Example scenario: a month of work that generates 1,000,000 input tokens, 1,000,000 output tokens, and 1,000,000 cache-read tokens on Claude Sonnet 5.5.

On CheapRelay: $0.8903 + $4.45 + $0.0778 = $5.4181, which at 7.2 yuan per dollar is ¥39.00912. At the official list price: $2.00 + $10.00 + $0.10 = $12.10, or ¥87.12. The difference is $6.6819 per million-token cycle, which is ¥48.11088. Ten identical cycles turn that into $66.819, or ¥481.1088.

Example scenario: a heavier, read-heavy month with 10,000,000 input tokens and 2,000,000 output tokens, no cache reads. On CheapRelay, 10 × $0.8903 = $8.903 and 2 × $4.45 = $8.90, so $17.803, or ¥128.1816. At list, 10 × $2.00 = $20 and 2 × $10.00 = $20, so $40, or ¥288. The saving is $22.197, which is ¥159.8184.

These are examples, not quotes. Your bill depends on your token mix, and the price list is where you check the current rates before you pay.

04 / Where Claude Sonnet 5.5 sits against the rest of the price list

It helps to know what you are paying for. Claude Sonnet 5.5 costs $0.8903 input, $4.45 output, and $0.0778 cache read. Claude Sonnet 5 costs $0.9764, $4.88, and $0.0861, so Sonnet 5.5 is lower by $0.0861 on input, $0.43 on output, and $0.0083 on cache reads. Claude Opus 5.5 costs $1.32, $6.24, and $0.0597: its input is $0.4297 higher and its output $1.79 higher, while its cache read is $0.0181 lower. Claude Opus 5 costs $1.66, $8.28, and $0.1222, higher on all three lines. Claude Haiku 4.5 costs $0.2917 input, $1.40 output, and $0.025 cache read, so it runs $0.5986 less on input and $3.05 less on output per million tokens.

Outside the Claude family, compare price list entries directly. Qwen3.7 Plus lists $0.1208 input, $0.3625 output, and $0.0361 cache read. MiMo V2.6 Pro lists $0.1736, $0.3472, and $0.002778. MiniMax M3 lists $0.1236, $0.4917, and $0.025. DeepSeek V4 Flash lists $0.2389, $0.9556, and $0.005556. Each sits below Sonnet 5.5 on input and output, and the cache-read figures compare as $0.002778 for MiMo V2.6 Pro, $0.005556 for DeepSeek V4 Flash, $0.025 for MiniMax M3, and $0.0361 for Qwen3.7 Plus. If your workload is light on tokens, those lines cost less; if you want a Claude model through the official Claude clients, Sonnet 5.5 is the one priced out here.

GPT rows carry official list prices too, and the same arithmetic applies. GPT-6.1 Sol lists $0.7375 output against an official $10.00, a gap of $9.2625 per million output tokens. CheapRelay publishes these numbers side by side so you can run the comparison yourself before committing.

05 / Your Claude Sonnet 5.5 setup on the first day

Claude models on CheapRelay work with the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. Point the client at the API base api.cheapapirelay.com, enter the model id claude-sonnet-5-5, and send a request. That is the whole path.

One rule saves time on day one: Claude models do not run through Cursor, Cherry Studio, Cline, Roo Code, OpenCode, an SDK, or cURL, and they are not OpenAI-compatible. Those OpenAI-compatible clients under /v1 — Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, cURL — are the route for GPT (including Codex), DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax, and the other models on the price list. Gemini is separate again: native protocol only, under /v1beta, through Gemini CLI.

Two limits are worth knowing before you script against a key. Each key allows 120 calls per minute, and each account allows 8 calls at once. If a request returns HTTP 402, the balance is at or below zero, so top up and try again. With prepaid billing, a $5 balance at $0.8903 per million input tokens covers 5 ÷ 0.8903 = 5.62 million input tokens before output is counted, so plan the top-up around the mix you actually send.

06 / Prepaid balance, refunds, and referral arithmetic

CheapRelay charges no monthly fee. You top up, the balance is prepaid, and it does not expire, so an unused $3 in January is still $3 in December. The refund window is 12 days from the top-up, covering the unused balance minus what was spent and the payment fee. Example scenario: you top up $5, spend $1.20, and the payment fee is $0.40, so the refund is $5.00 − $1.20 − $0.40 = $3.40.

The referral program is also plain arithmetic. A directly invited user who pays $10 puts 20% × $10 = $2 into your withdrawable rewards balance. A user invited at the second level who pays $10 puts 6% × $10 = $0.60 there. The smallest withdrawal is ¥20, paid through Alipay.

Put the pieces together for one example: a team that sends Claude Sonnet 5.5 ten million input tokens and two million output tokens in a month pays $17.803 on CheapRelay, against $40 at the official list price, with no monthly fee attached to either route. Check the price list, pick the client that matches your model, and let the token counts decide.