BestRelay and GPT-6 Astra: pay-as-you-go pricing, plainly

BestRelay forwards each request to the model you wrote in the name

Here is an example day at your own desk, with GPT-6 Astra doing the work. You open with a request that asks for long reasoning: a draft to be checked, notes to be summarised, a decision to be argued through from both sides. You name GPT-6 Astra, and BestRelay forwards that request to the provider under the name you wrote. The model named is the model that answers, and nothing is quietly swapped for a cheaper tier. That sentence is the point on which the rest of this article rests, and the rest is mostly arithmetic in its favour.

For example, a day of 500,000 input tokens and 100,000 output tokens costs $0.78 on BestRelay. 22 such days cost $17.11. These are examples rather than forecasts; they are what the rate card produces when the token counts are these and the model is GPT-6 Astra.

At the official list price, the same day is $10.00 and the 22 days are $220.00. The difference on one day is $9.22; across 22 days it is $202.89. The work does not change, the model does not change, and the invoice at the end of it changes quite a lot.

BestRelay is an independent API relay. It keeps a curated line-up of 20 public models, each listed under its own name with four prices: input, output, cache write and cache read. When a provider retires a model id, BestRelay delists it. You choose from a short, deliberate list, and you see the rate card before you spend anything.

Cache reads make a long reasoning day cheaper on BestRelay

Long reasoning work tends to repeat itself: the same brief, the same source material, the same standing instructions, carried from one call to the next. That repetition is where the cache read price matters, and BestRelay lists it plainly beside the other three rates for every model in the line-up.

For example, on GPT-6 Astra, a day of 500,000 input tokens and 100,000 output tokens costs $0.78; with 350,000 of the input read from cache, the day costs $0.54. At the official list price, the same uncached day is $10.00 and the cached day is $6.85. The difference on the day is $9.22.

The same arithmetic, as an example, with Claude Sonnet 5.5: a day of 500,000 input and 100,000 output tokens costs $1.40 on BestRelay, 22 such days cost $30.71, and 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 as an example with GPT-6.1 Sol: 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.

BestRelay bills each call after it ends, from a prepaid balance

Here is the shape of the bill in an example. You top up, you work, and BestRelay charges each call once, after it ends, from the tokens the provider reports, at the rate card prices. Nothing is held in advance. There is no monthly fee, no subscription and no minimum spend; the balance simply moves as calls are made.

The top-up amounts, through Stripe checkout, are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. For example, a $13.89 top-up is a modest way to begin, and the amount is yours to choose from that list. Each new account opens with $0.35 of credit, which is enough to make a few calls and read the usage records before committing anything further. Every call appears in the usage records with its model, tokens and charge.

The failure cases follow a plain rule. Requests the provider rejects are recorded at zero, and when the provider cannot be reached nothing is charged. You are not billed for a call that did not happen.

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. BestRelay does not leave the small print to do the work; the terms are stated plainly, and the balance is the record.

Two settings and a key: how the desk example is wired

Setting up BestRelay takes two settings: the address and a key created under API keys in the dashboard. The details are in the docs, and they are not reproduced here.

The client you use decides which models you can name. Claude models are served 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. Codex, like those other tools, pairs with GPT, DeepSeek, Kimi, GLM, Qwen and Grok. Gemini models use their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible. Knowing this before you start saves an afternoon of mismatched settings.

The limits are modest and stated: 120 requests a minute on one key, and 8 calls in flight on one account. An account can hold several keys, and each can be voided at any time, so a key that has travelled too far can be retired without closing the account.

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. The desk in this example is yours, the text you type is yours, and the record of what you spent is yours as well.

A curated line-up of 20 models, and a way to share the cost

BestRelay's line-up is curated: 20 public models, each listed under its own name with four prices — input, output, cache write and cache read. Among them sit GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5, Claude Fable 5.1, Claude Haiku 4.5, DeepSeek V4 Pro, Gemini 4 Argon, Gemini 3.8 Flash, Kimi K3, Qwen3.8 Max, Grok 4.7 and GLM-5.3, among others. You name the model; BestRelay forwards the request to the provider under that name. When a provider retires a model id, BestRelay delists it rather than leaving a stale entry behind.

There is also a referral scheme, and it is worth reading closely. 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 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.

So the example day at your desk ends the way it began: with the model you named doing the answering, a usage record showing the tokens and the charge, and a balance that reflects exactly what was used. If you would like to see how that reads on your own work, BestRelay is ready when you are.