Gemini CLI on BestRelay: reading the API proxy rate card

BestRelay sends each Gemini CLI request to the model name you wrote

BestRelay is an independent API relay, and its promise is narrow and plain: the model you name is the model that answers. Each request is forwarded to the model provider under the model name the caller wrote, so nothing is quietly swapped for a cheaper tier behind the scenes. For Gemini work the route is the native protocol, through Gemini CLI or direct generateContent calls; Gemini models are not OpenAI-compatible, and BestRelay does not pretend otherwise.

The line-up is curated. Twenty public models sit on it, each listed under its own name with four prices beside it — input, output, cache write and cache read — so a reader can read the card line by line rather than guess at a tier. When a provider retires a model id, BestRelay delists it, which keeps the card honest about what can actually be called. Gemini CLI users get the Gemini rows: Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash and Gemini 4 Argon.

That discipline matters most in a relay, where the temptation is to route around what was asked for. BestRelay takes the other course. You write the model name, the provider receives that name, and the answer comes back from that model. The record of the call — its model, tokens and charge — appears in the usage records, so the claim can be checked after the fact instead of taken on trust. For a reader who has met a service that quietly substituted a smaller model, this is the point worth weighing.

The Gemini rows of the BestRelay rate card, read line by line

Every figure on the card is quoted per 1,000,000 tokens, and every model carries the same four columns: input, output, cache write and cache read. The Gemini section is short enough to read in a minute, which is the point of publishing it in full.

• Gemini 2.5 Flash — input $0.2778, output $2.08, cache write $0.2778, cache read $0.0264.
• Gemini 3.1 Pro — input $1.74, output $9.93, cache write $1.74, cache read $0.1389.
• Gemini 3.8 Flash — input $1.25, output $5.56, cache write $1.25, cache read $0.1236.
• Gemini 4 Argon — input $1.67, output $8.06, cache write $1.67, cache read $0.0847.

Read across the four lines and the shape is consistent. In each Gemini row the cache write price matches the input price, while cache read sits lower, which is the usual reason to arrange for repeated material to be read from cache rather than sent again. Gemini 2.5 Flash is the light entry in the group, with input at $0.2778 and cache read at $0.0264; Gemini 3.1 Pro is the heavier one, at $1.74 input and $9.93 output.

There is no hidden column. Gemini 3.8 Flash shows $1.25 input, $5.56 output, $1.25 cache write and $0.1236 cache read, and Gemini 4 Argon shows $1.67 input, $8.06 output, $1.67 cache write and $0.0847 cache read. Since Gemini models run only on their native protocol, through Gemini CLI or direct generateContent calls, the card has to be legible to people reading the documentation for that protocol rather than a compatibility shim. BestRelay publishes the numbers as they stand, and the usage records show the model, tokens and charge for every call.

BestRelay charges a prepaid balance after each call ends

BestRelay runs on a prepaid, pay-as-you-go balance. There is no monthly fee, no subscription and no minimum spend. Each call is charged once, after it ends, using the tokens the provider reports and the rate card prices. Nothing is held in advance, so an account carries no standing commitment beyond the credit already in it.

The unhappy paths are handled plainly. A request the provider rejects is recorded at zero, and when the provider cannot be reached, nothing is charged. A new account opens with $0.35 of credit, enough to begin calling while you read the card.

Top-ups run through Stripe checkout, and the amounts are fixed: $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. 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.

Referrals are paid on the top-ups of people who joined through your link: 18% for you, and 4% for the people they invited in turn. As an example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556. Commission can be moved into the balance, or withdrawn to Alipay from $20.83.

Connecting Gemini CLI to BestRelay takes an address and a key

A client needs two settings: the address and a key created under API keys in the dashboard. The details live in the BestRelay docs, and this article prints neither, which is the correct habit for a page others may copy. Point Gemini CLI at the native protocol, supply the key, and the model name in your request is the model that answers.

The limits are stated in plain numbers. One key can carry 120 requests a minute, and one account can have 8 calls in flight. An account may hold several keys, and any of them can be voided at any time, so a key that travels too far can be retired in a moment. Every call appears in the usage records with its model, tokens and charge.

Other clients have their own rules. 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.

On privacy the position is specific rather than sweeping. 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.

What the BestRelay numbers look like in ordinary use

A rate card is easier to judge with a day of traffic behind it. 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; the difference on the day is $0.60 and on the 22 days $13.29.

GPT-6.1 Sol, as an example, runs smaller still: a day of 500,000 input and 100,000 output tokens costs $0.22 on BestRelay, 22 such days cost $4.74, and 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; the difference on the day is $1.78 and on the 22 days $39.26.

The free credit alone covers about 1.6 million GPT-6.1 Sol input tokens, which is enough to see how the card behaves before topping up. BestRelay keeps the arrangement simple throughout: name the model, read the price, pay for what the provider reports. If you would like to try it, an account with $0.35 of credit is a modest place to begin.