BestRelay with Claude Code: an API relay priced by the token

BestRelay forwards each request to the model you wrote

BestRelay is an independent API relay, and the promise behind it is plain: the model you name is the model that answers. Each request is forwarded to the provider under the model name the caller wrote, and nothing is quietly swapped for a cheaper tier. For a company, that is not a small detail. It means the name in your code, the name in the request and the name in the usage record are the same name, and the answer came from that model.

Here is an example of a sample day at your own desk. You open BestRelay in the morning with a prepaid balance, send a batch of drafting to Claude Sonnet 5.5, and read the charge afterwards. For example, a day of 500,000 input and 100,000 output tokens costs $1.40 on BestRelay, and 22 such days cost $30.71. With 350,000 of the input read from cache, the same day costs $0.94. At the official list price the same day is $2.00, the 22 days $44.00 and the cached day $1.33. The difference on the day is $0.60, and on the 22 days $13.29.

The same shape holds for a different line. For example, a day of 500,000 input and 100,000 output tokens on GPT-6.1 Sol costs $0.22 on BestRelay, and 22 such days cost $4.74, while with 350,000 of the input read from cache the day costs $0.16. At the official list price the same day is $2.00, the 22 days $44.00 and the cached day $1.33, so the difference on the day is $1.78 and on the 22 days $39.26. Name the model, send the work, read the bill.

The BestRelay line-up is 20 public models, each priced under its own name

BestRelay does not present one anonymous pool of capacity. It presents a curated line-up of 20 public models, and each one is listed under its own name with four prices: input, output, cache write and cache read. The prices are per 1,000,000 tokens, and they are the prices you are charged. A colleague can open the list and check a figure without asking anyone, which is a modest courtesy that pays for itself during a busy week.

A few entries from the sheet:

• Claude Haiku 4.5 is $0.4167 input, $2.01 output, $0.4861 cache write and $0.0333 cache read.
• Claude Opus 5.5 is $1.74 input and $8.13 output, with cache write at $2.01 and cache read at $0.0778.
• Claude Fable 5.1 is $4.10 input, $20.42 output, $5.00 cache write and $0.2083 cache read.
• GPT-6.1 Sol is $0.2083 input, $1.11 output and $0.0431 cache read.
• Gemini 2.5 Flash is $0.2778 input, $2.08 output, $0.2778 cache write and $0.0264 cache read.
• DeepSeek V4 Pro is $1.53 input, $4.58 output and $0.0514 cache read.
• Grok 4.6 and GLM-5.3 both sit at $0.625 input and $1.94 output, with cache read at $0.1389.

Two housekeeping details matter. The line-up is curated rather than accidental: twenty entries, each named, each priced in the open. And when a provider retires a model id, BestRelay delists it, so the names on the list are the names that can still answer. If a model disappears, it is because the provider withdrew it, not because BestRelay quietly moved it sideways. That is what makes the model you name the model that answers, and what makes the bill readable.

BestRelay bills a prepaid balance, once per finished call

BestRelay uses a prepaid pay-as-you-go balance. There is no monthly fee, and each call is charged once, after it ends, from the tokens the provider reports, at the rate card prices. Nothing is held in advance, so the balance you see is the balance you have, and a quiet afternoon of experiments does not leave a surprise on the statement.

Failure has its own rule, which is the rule you would want on a bill. Requests the provider rejects are recorded at zero, and when the provider cannot be reached, nothing is charged. A rejected prompt therefore costs nothing to try, and an outage on the provider side does not drain the balance.

Each new account opens with $0.35 of credit. For example, the free credit alone covers about 1.6 million GPT-6.1 Sol input tokens, which is a tidy way to try the service before you top up. Top-ups run through Stripe checkout, in these amounts: $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. You choose the size that suits the month you can actually see.

Unused credit can be refunded on request within 13 days of the top-up, less what was consumed and any payment-processor fees. Gift credit and commission cannot be refunded, which is worth knowing before you treat referral earnings as spendable cash. Referrals are a separate line: you earn 18% of the top-ups of people who joined through your link, and 4% of the top-ups of the people they invited. Commission can be moved into the balance, or withdrawn to Alipay from $20.83. For example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556.

Clients and connections follow stated rules on BestRelay

BestRelay is a relay, and it behaves like one: it forwards, it records, and it charges what the rate card says. Which client you may use follows from the model you want to call, and those rules are worth reading before a colleague wires anything up.

Claude models are served only to the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. They do not work in Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL. Those tools, like Codex, pair only with GPT, DeepSeek, Kimi, GLM, Qwen and Grok. Gemini models use only their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible.

Connecting a client takes two settings: the address, and a key created under API keys in the dashboard. The details are in the docs. We do not print them here, and you would not paste them into a shared page either.

For a company, this matters in a practical way. Your team can standardise on a handful of named models, give each developer their own key, and know that the model written in the request is the model that did the work. If a colleague needs Claude Fable 5.1 rather than Gemini 3.8 Flash, that is a change of name in the request, not a change of supplier behind the curtain. The same is true when you compare Gemini 3.1 Pro at $1.74 input and $9.93 output with Claude Sonnet 5 at $1.32 input and $6.67 output: you are choosing between named lines with published prices, not guessing at a tier.

Keys, usage records and limits keep the sample day tidy

An account can hold several keys, each can be voided at any time, and every call appears in the usage records with its model, tokens and charge. If a key goes astray, you revoke it and move on; the record of what it did stays readable, which is the difference between an audit trail and a guessing game.

There are limits, and they are stated rather than implied: 120 requests a minute on one key, and 8 calls in flight on one account. For a desk-sized team, that is a ceiling you can plan around, and for a larger one it is a number you can plan around before the afternoon gets awkward.

Privacy is narrow by design. The user-written text of requests is kept for 30 days, to review abuse and settle billing questions. System prompts and model replies are not stored.

Put the sample day back together. You arrive with a prepaid balance, choose a named model, send the request, and watch the charge land once the call ends. The usage record names the model, the tokens and the cost. Nothing is held back in advance, nothing is charged when the provider rejects the call, and nothing is quietly swapped for a cheaper tier. If you would like to see how that reads on your own work, BestRelay is worth a try.