BestRelay setup for Gemini CLI, one clear API relay rate card

BestRelay sends each request to the model name you wrote

BestRelay is an independent API relay, and its central promise is plain: it forwards each request to the model provider under the model name the caller wrote. The model named is the model that answers, and nothing is quietly swapped for another tier behind the scenes. That is the point the rest of this article supports, and it is worth reading against the rate card rather than in the abstract.

The line-up is curated. BestRelay lists 20 public models, each under its own name, each with four prices: input, output, cache write and cache read. When a provider retires a model id, BestRelay delists it rather than leaving a stale entry in place. So the card you read is the card you are billed from, and a retired name does not linger as a trap.

Take one line of the card as an example. Claude Fable 5 is priced at $4.17 input, $20.97 output, $5.14 cache write and $0.3472 cache read per 1,000,000 tokens, against an official list price of $10 input, $50 output and $1 cache read. Claude Haiku 4.5 sits much nearer the floor of the card: $0.4167 input, $2.01 output, $0.4861 cache write and $0.0333 cache read, with an official list price of $1 input, $5 output and $0.1 cache read.

Read that way, BestRelay is not a bundle of vague promises but a list you can check before you send anything. The name in your request, the name in the listing, the prices attached to it: all three line up, and the same discipline applies to every other entry on the card.

BestRelay bills each call once, after it finishes

BestRelay runs on a prepaid, pay-as-you-go balance. There is no monthly fee, no subscription and no minimum spend, and nothing is held in advance. Each call is charged once, after it ends, using the token counts the provider reports and the rate card prices for that model.

Because the charge lands only when the call is finished, a long-running request does not quietly draw down your balance while it is still working. The balance moves at the end, and the usage records show the model, the tokens and the charge for every call, so the figures on the card can be checked after the fact rather than merely trusted.

Failure is priced honestly, too. A request the provider rejects is recorded at zero, and when the provider cannot be reached at all, nothing is charged. You are not asked to pay for a call that produced no answer.

Every new BestRelay account opens with $0.35 of credit. As an example, the free credit alone covers about 1.6 million GPT-6.1 Sol input tokens, which is enough to watch the reporting and the charges behave before you add funds.

Top-ups are handled through Stripe checkout, in these amounts: $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. The choice is a matter of how much runway you want, not of which tier you are entitled to; the model you name is still the model that answers. And because nothing is held in advance, an unused balance simply stays on the account until you spend it or request a refund inside the window described below.

Refunds and referral commission follow stated windows and percentages

Unused credit can be refunded on request within 13 days of the top-up, less what was consumed and any payment-processor fees. Read in order, the rule is narrow and clear: it covers unused top-up credit inside a 13-day window, and it names its two exclusions without hedging. Gift credit and commission cannot be refunded.

Referrals work on two levels. You receive 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 your balance, or withdrawn to Alipay from $20.83. Nothing about that arrangement depends on a recurring plan, because there is no monthly fee, no subscription and no minimum spend in the first place.

As an example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556. That is the shape of the scheme in a single line: a share of what your invitees spend, and a smaller share of what their invitees spend.

BestRelay keeps the whole arrangement inside one prepaid, pay-as-you-go balance. Commission can sit in the account, be moved into spendable credit, or be withdrawn to Alipay once it reaches $20.83, while a refund request inside 13 days returns the unused part of a top-up less consumption and processor fees. The rules are short, and they are printed rather than implied.

The BestRelay rate card, read line by line

Every model on the BestRelay card is listed with the same four figures per 1,000,000 tokens: input, output, cache write and cache read. Where the card gives an official list price, it is printed alongside.

Claude Opus 5.5 is $1.74 input, $8.13 output, $2.01 cache write and $0.0778 cache read, against an official list price of $4 input, $20 output and $0.2 cache read. Claude Sonnet 5 is $1.32 input, $6.67 output, $1.67 cache write and $0.1167 cache read. On the GPT side, GPT-6.1 Sol is $0.2083 input, $1.11 output and $0.0431 cache read, with an official list price of $2 input, $10 output and $0.1 cache read. Gemini 3.1 Pro is $1.74 input, $9.93 output, $1.74 cache write and $0.1389 cache read. DeepSeek V4 Pro is $1.53 input, $4.58 output and $0.0514 cache read.

Worked examples make the card concrete. Take Claude Sonnet 5.5 as an 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 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, a difference of $0.60 on the day and $13.29 over the 22 days.

Take GPT-6.1 Sol as another example: a day of 500,000 input and 100,000 output tokens costs $0.22 on BestRelay, and 22 such days cost $4.74. With 350,000 of the input read from cache, the day costs $0.16, against $2.00, $44.00 and $1.33 at the official list price, a difference of $1.78 on the day and $39.26 over the 22 days. The free credit alone covers about 1.6 million GPT-6.1 Sol input tokens.

Those examples are arithmetic on the published card, not a claim about anyone's traffic. The card itself is the instrument: check the four prices for the model you intend to name, and you can price the call before you make it.

Which client you use decides which model can answer

BestRelay serves Claude models only to the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. Other tools, such as Codex, pair only with GPT, DeepSeek, Kimi, GLM, Qwen and Grok models. Gemini models use only their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible. In every supported pairing, the model named in the request is the model that answers.

Connecting a client takes two settings: the address and a key created under API keys in the dashboard. The details are in the BestRelay docs. Nothing else is required, and no configuration is guessed on your behalf.

The account carries stated limits: 120 requests a minute on one key, and 8 calls in flight on one account. Keys are plural by design. An account can hold several, and any one of them can be voided at any time, which is useful when a key travels further than you intended. Every call appears in the usage records with its model, tokens and charge.

On privacy, BestRelay keeps the user-written text of requests for 30 days, to review abuse and settle billing questions. System prompts and model replies are not stored. That is the retention policy in full, stated plainly.

So the reading ends where it began: a curated card of 20 named models, a prepaid balance charged once per finished call, and the model you name answering your request. Give BestRelay a try with the $0.35 of opening credit and check the charges against the card as they arrive.