What does Claude Haiku 4.5 pricing look like on CheapRelay?

01 / Claude Haiku 4.5 pricing starts with three numbers

Claude Haiku 4.5 pricing gets easier to judge when every line of the bill is visible. On the CheapRelay price list, the model id is claude-haiku-4-5-20251001. Prices are per 1,000,000 real tokens:

• ■ Input: $0.2917
• ■ Output: $1.40
• ■ Cache read: $0.025

That is the whole price structure. There is no monthly fee. You add funds to a prepaid balance, and the balance does not expire. CheapRelay charges per real token used, so a small test does not require a long commitment. Every new account also receives $0.09 in sign-up credit that can be used on any model.

The official list price for the same Claude Haiku 4.5 rates is $1.00 for input, $5.00 for output, and $0.10 for cache read per 1,000,000 tokens. CheapRelay uses the stated comparison rate of 7.2 yuan per US dollar. Here is the arithmetic:

• ■ Input: $1.00 − $0.2917 = $0.7083, or 70.83% below the official list price.
• ■ Output: $5.00 − $1.40 = $3.60, or 72% below the official list price.
• ■ Cache read: $0.10 − $0.025 = $0.075, or 75% below the official list price.

Converted at 7.2 yuan per dollar, the CheapRelay rates are 2.10024 yuan for input, 10.08 yuan for output, and 0.18 yuan for cache read per million tokens. The corresponding official list figures are 7.2 yuan, 36 yuan, and 0.72 yuan. The price gap is not a vague promise; it is visible in each line.

02 / One million tokens, calculated from both sides

Here is an example using 1,000,000 input tokens and 1,000,000 output tokens, with no cache reads. The CheapRelay calculation is:

• ■ Input: 1,000,000 ÷ 1,000,000 × $0.2917 = $0.2917
• ■ Output: 1,000,000 ÷ 1,000,000 × $1.40 = $1.40
• ■ Total: $0.2917 + $1.40 = $1.6917

At the official list price, the same example would cost:

• ■ Input: 1,000,000 ÷ 1,000,000 × $1.00 = $1.00
• ■ Output: 1,000,000 ÷ 1,000,000 × $5.00 = $5.00
• ■ Total: $1.00 + $5.00 = $6.00

The difference is $6.00 − $1.6917 = $4.3083. Divide that by $6.00 and the saving is 71.805%, or about 71.81%. In yuan, $1.6917 × 7.2 = 12.18024 yuan, compared with $6.00 × 7.2 = 43.20 yuan. The saving is 31.01976 yuan.

Now add a cache read to the example. Suppose a request consumes 100,000 input tokens, produces 20,000 output tokens, and reads 500,000 cached tokens. On CheapRelay:

• ■ Input: 0.1 × $0.2917 = $0.02917
• ■ Output: 0.02 × $1.40 = $0.028
• ■ Cache read: 0.5 × $0.025 = $0.0125
• ■ Total: $0.02917 + $0.028 + $0.0125 = $0.06967

The same token mix at official list pricing is 0.1 × $1.00 + 0.02 × $5.00 + 0.5 × $0.10 = $1.125. CheapRelay saves $1.05533 on this example, which is 93.81% of the official total. That is why the cache-read line deserves attention: it is $0.025 on CheapRelay, not $0.10 on the official list.

03 / What a larger run costs in plain numbers

Small prices become easier to understand when multiplied by a realistic workload. Here is another example: 100 calls, each using the same 100,000 input tokens, 20,000 output tokens, and 500,000 cache reads.

• ■ One call on CheapRelay: $0.06967
• ■ 100 calls on CheapRelay: 100 × $0.06967 = $6.967
• ■ One call at official list price: $1.125
• ■ 100 calls at official list price: 100 × $1.125 = $112.50
• ■ Difference: $112.50 − $6.967 = $105.533

At 7.2 yuan per dollar, those totals become 50.1624 yuan on CheapRelay and 810 yuan at official list pricing. The example saves 759.8376 yuan.

Scale it once more, as an example rather than a promise. For 1,000 identical calls:

• ■ CheapRelay: 1,000 × $0.06967 = $69.67, or 501.624 yuan.
• ■ Official list price: 1,000 × $1.125 = $1,125, or 8,100 yuan.
• ■ Difference: $1,125 − $69.67 = $1,055.33, or 7,598.376 yuan.

That gap comes from three published rates, not from an invented discount. CheapRelay charges $0.2917 for input, $1.40 for output, and $0.025 for cache reads per million tokens. The official list rates are $1.00, $5.00, and $0.10. You can reproduce every total above with a calculator.

Prepaid billing also keeps the workflow direct. If the balance reaches zero or below, an HTTP 402 response tells you that funds need attention. There is still no monthly fee. Your balance does not expire, and unused balance from a top-up can be refunded within 12 days, minus what has been spent and the payment fee.

04 / Where Haiku 4.5 sits against other priced models

Haiku 4.5 is not the only option on the CheapRelay price list, so compare the actual figures before choosing. On output, Claude Haiku 4.5 is $1.40 per million tokens. Gemini 2.5 Flash is $1.42, so Haiku is $0.02 lower per million output tokens. GLM-5.3 is $1.38, which is $0.02 lower than Haiku. These are narrow differences, and the correct choice depends on the workload.

Compare Haiku with GPT-6.1 Sol and the arithmetic changes. GPT-6.1 Sol output is $0.7375 per million tokens. Haiku output is $1.40 − $0.7375 = $0.6625 higher. That makes Haiku output 89.83% higher than GPT-6.1 Sol output. On input, GPT-6.1 Sol is $0.1486 and Haiku is $0.2917, a difference of $0.1431 per million tokens.

The official list comparison for Haiku is more direct because both figures are published for the same model. Its $1.40 output rate is $3.60 below the $5.00 official list rate. Its $0.2917 input rate is $0.7083 below $1.00. Its $0.025 cache-read rate is $0.075 below $0.10.

CheapRelay also lists MiniMax M3 at $0.1236 input, $0.4917 output, and $0.025 cache read per million tokens; Qwen3.7 Plus at $0.1208 input, $0.3625 output, and $0.0361 cache read; and DeepSeek V4 Flash at $0.2389 input, $0.9556 output, and $0.005556 cache read. Comparing those three published rates with Haiku’s $0.2917 input, $1.40 output, and $0.025 cache read lets you match the bill to the model rather than trusting a slogan.

05 / Connect the right client and keep the setup simple

Claude Haiku 4.5 runs through the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. Set the model id to claude-haiku-4-5-20251001. The API base for CheapRelay is https://api.cheapapirelay.com.

Keep the client pairing clear. Claude models use the official Claude Code and Claude Desktop clients. Gemini models use their native protocol under /v1beta through Gemini CLI. GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax, and the other OpenAI-compatible models use an OpenAI-compatible client under /v1, such as Cursor, Cherry Studio, Cline, Roo Code, OpenCode, the OpenAI SDK, or cURL. Claude and Gemini are not OpenAI-compatible routes.

CheapRelay allows 120 calls per minute per key and 8 concurrent calls per account. Those limits are stated plainly, so you can plan a workload without discovering hidden constraints later. Billing also has a simple referral rule: a directly invited user generates a 20% reward on what they pay, while the second level generates 6%. Rewards go into a withdrawable balance, and the smallest withdrawal is ¥20 via Alipay.

For a buyer comparing cost and setup, the final arithmetic is short. Claude Haiku 4.5 costs $0.2917 per million input tokens, $1.40 per million output tokens, and $0.025 per million cache reads on CheapRelay. The official list rates are $1.00, $5.00, and $0.10. In the 100-call example above, the difference is $105.533, or 759.8376 yuan at the stated conversion rate. You use the official Claude clients, pay per real token from a prepaid balance, and keep every published rate visible.