BestRelay API relay: Gemini 4 Argon at $1.67 input pricing

BestRelay forwards each request to the model you name, without substitution

You open your editor on a quiet Tuesday and decide to run a check on Gemini 4 Argon. The prompt is a routine refactor: a few hundred lines of TypeScript that need stricter types and a lighter dependency graph. You have an account with BestRelay, an independent API relay that forwards each request to the model provider under the model name the caller wrote, so the model named is the model that answers and nothing is quietly swapped for a cheaper tier. The line-up is curated, 20 public models, each listed under its own name with four prices (input, output, cache write, cache read). When a provider retires a model id, BestRelay delists it, so the list you read is the list that answers.

You created your key under API keys in the dashboard, pasted the address and the key into Gemini CLI, and the call returned a clean diff. That moment confirms what the service promises. Money moves in a way you can follow: BestRelay runs on prepaid pay-as-you-go balance. No monthly fee, no subscription, no minimum spend. Each call is charged once, after it ends, from the tokens the provider reports, at the rate card prices. Nothing is held in advance. Your account opened with $0.35 of free credit, which covers about 1.6 million GPT-6.1 Sol input tokens, as an example. When you need more, you top up through Stripe checkout in fixed amounts: $1.39, $6.94, $13.89, $27.78, $69.44, $138.89.

Refunds are plain as well. 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. You keep a note of this, because it tells you what to do with a balance you might not spend in full.

Gemini models on BestRelay use only their native protocol, through Gemini CLI

You are careful about client compatibility, because the wrong pairing wastes an afternoon. 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. You once tried to point a generic OpenAI-compatible client at the endpoint and met a protocol error; switching to Gemini CLI resolved the matter at once.

Connecting a client takes two settings, the address and a key created under API keys in the dashboard, and the details sit in the docs. The pace of your work is governed by clear limits: 120 requests a minute on one key, and 8 calls in flight on one account. You keep your own key for experiments and a separate one for the project, so a stray script cannot crowd out the work you care about. An account can hold several keys, and each can be voided at any time. That single property is enough for you to keep an old key around without worrying about it.

Every call appears in the usage records with its model, tokens and charge, which is where you check a figure whenever a request feels heavier than expected. On the subject of stored material, the arrangement is equally plain: the user-written text of requests is kept for 30 days to review abuse and settle billing questions, while system prompts and model replies are not stored. Your prompt text sits in the record; the model side of the exchange does not.

BestRelay charges once per finished call, and records failures at zero

Mid-morning you restart the provider connection twice, and the relay behaves the way you have been told it behaves. Requests the provider rejects are recorded at zero, and when the provider cannot be reached nothing is charged. So the two attempts that never produced a reply cost nothing, and the call that completed is the only one that appears as a charge in your usage records. Nothing is held in advance either: the balance is drawn down after a call ends, from the tokens the provider reports, at the rate card prices.

You also keep an eye on the referral side of the account, because a link you share occasionally earns something. Referrals pay 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. As an example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556. You leave the commission where it sits until it clears the withdrawal threshold, then move it into the balance for the coming week. Gift credit and commission cannot be refunded, so the sensible habit is to treat them as working credit rather than cash.

The daily rhythm is uncomplicated: check the balance, run the calls, read the records. Each entry names the model, the tokens and the charge, so you can tell at a glance whether the day matched the rate card or whether something needs adjusting. When a provider retires a model id, BestRelay delists it, and when a call fails, the ledger stays honest. Both habits make the same point, and the point is the reason you stay with BestRelay.

One desk, one address, several models, and figures you can check

Here is an example day, plainly labelled as an example. You work at your own desk with Gemini 4 Argon as your main model, running a refactor in the morning and a documentation pass in the afternoon. The Gemini 4 Argon rate card reads input $1.67, output $8.06, cache write $1.67, cache read $0.0847. As an example, a day of 500,000 input and 100,000 output tokens costs $1.64 on BestRelay, 22 such days cost $36.06; with 350,000 of the input read from cache the day costs $1.09. You notice how small the cache read rate sits beside the input rate, so you arrange your prompts to reuse context instead of resending it.

Later you try a second model for a different kind of task, changing only the model name in the request. Claude Sonnet 5.5, as an example: a day of 500,000 input and 100,000 output tokens costs $1.40 on BestRelay, 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. And GPT-6.1 Sol, as an example: a day of 500,000 input and 100,000 output tokens costs $0.22 on BestRelay, 22 such days cost $4.74; 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.

You do not need a second account, a second address or a second key to move between them. The curated line-up of 20 public models sits behind the same connection, and the model you name is the model that answers.

BestRelay keeps the mechanics quiet so your attention stays on the work

By the end of the example day, the ticket is closed and the usage page holds the whole story: the models you asked for, the tokens each call used, the charge each call produced, and the failed attempts sitting at zero. You had topped up $13.89 earlier in the day and drawn on the $0.35 of credit your account opened with, and the balance still covers what comes next. If you ever wanted the unused part back, you would ask within 13 days of the top-up, less what was consumed and any payment-processor fees.

You close the laptop with the address and key still in the client, ready for tomorrow. The relay does not ask for attention, and it does not substitute anything behind your back. It forwards each request to the model provider under the model name the caller wrote, prices the tokens the provider reports, and writes the result into records you can read at any time. The native protocol is handled, the failures cost nothing, and the line-up is curated and current. Try BestRelay, name the model you want, and see how the desk feels when the machinery in between stays out of the way.