BestRelay setup with Claude Code: one API relay, named models

401, 402 and 429 are three ordinary answers to three desk-day questions

Here is a plainly labelled example of a desk day with BestRelay, written to you as you rather than to an audience. You arrive with a question, a client and a key, and three numbers turn up more often than you would expect: 401, 402 and 429. None of them is a comment on your taste in models. Each is a short technical answer to a practical situation, and BestRelay is easier to use once you know which situation produced the code.

401 arrives when the key you presented was not accepted. A BestRelay client needs two settings: the address, and a key created under API keys in the dashboard. The key identifies you; the address says where the request should go. If the key is mistyped, or does not belong to your account, you get an authentication failure rather than a reply. Keys can be voided at any time and an account can hold several of them, so replacing a suspect key is a small repair rather than a crisis. The details live in the BestRelay documentation, and reading them is the sensible move before changing anything else.

402 is the payment-required answer, and it belongs to the balance rather than to the key. BestRelay is prepaid and pay-as-you-go: you hold a balance, calls draw it down, and when the balance cannot cover the call, the request stops with a payment-required code. You will meet it if you have spent everything and forgotten to top up.

429 is the rate-limit answer. On BestRelay, one key is limited to 120 requests a minute, and one account may have 8 calls in flight. Exceed either and the service asks you to slow down. In our example day, the cure is simply to wait a moment and resend, not to change the model or rewrite the prompt. These three codes are short, and each one points at a specific fix.

BestRelay forwards the model name you wrote, without substituting a cheaper tier

The promise behind every BestRelay price is plain: the model you name is the model that answers. BestRelay is an independent API relay, and it forwards each request to the model provider under the model name the caller wrote. Nothing is quietly swapped for a cheaper tier on the way through. If your script asks for Claude Opus 5.5, Claude Opus 5.5 is what runs, and the usage records will say so.

The line-up is curated rather than sprawling: 20 public models, each listed under its own name with four prices, covering input, output, cache write and cache read. Among them sit Claude Fable 5, Claude Fable 5.1, Claude Haiku 4.5, Claude Opus 5, Claude Opus 5.5, Claude Sonnet 5, Claude Sonnet 5.5, DeepSeek V4 Pro, Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash, Gemini 4 Argon, Kimi K3, GPT-5.6 Sol, GPT-6 Astra, GPT-6.1 Sol, Qwen3.8 Max, Grok 4.6, Grok 4.7 and GLM-5.3. When a provider retires a model id, BestRelay delists it, so the list does not accumulate names that quietly fail.

Per 1,000,000 tokens, Claude Sonnet 5.5 is listed at $1.39 input, $7.01 output, $1.74 cache write and $0.0944 cache read. GPT-6.1 Sol is listed at $0.2083 input, $1.11 output and $0.0431 cache read. Claude Haiku 4.5 is $0.4167 input, $2.01 output, $0.4861 cache write and $0.0333 cache read. Each model carries its own rate card, and the model you pick decides what you pay per call. That is the whole idea in one sentence, and it is why the curated list is worth reading closely before you write anything.

Prepaid pay-as-you-go billing explains most of what a 402 is asking for

BestRelay bills from 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, from the tokens the provider reports, at the rate card prices for the model that answered. Nothing is held in advance, so an unused call costs you nothing, and a long call is charged once rather than in instalments.

Failed calls are handled with the same restraint. Requests the provider rejects are recorded at zero, and when the provider cannot be reached, nothing is charged. A relay that quietly billed for its own outages would be hard to trust; this one does not do it.

Every new account opens with $0.35 of credit, which is enough to run a small test and watch the usage records arrive with the model, tokens and charge for each call. Top-ups go through Stripe checkout in fixed amounts: $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 pay too. 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. For 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. On the privacy side, the text you write is kept for 30 days so abuse can be reviewed and billing questions settled; system prompts and model replies are not stored.

A sample desk day with GPT-6.1 Sol and Claude Sonnet 5.5 on BestRelay

Consider this example. You spend a day running Claude Sonnet 5.5 over 500,000 input tokens and 100,000 output tokens, and the cost on BestRelay is $1.40. Twenty-two such days come to $30.71. Now take a day in which 350,000 of the input is read from cache instead: 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, so the difference on the day is $0.60 and on the 22 days $13.29. Those are examples, not quotations, and they show how the rate card behaves in ordinary use.

Now the same arithmetic for GPT-6.1 Sol, again as an 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. At the official list price the same day is $2.00, the 22 days $44.00 and the cached day $1.33, which puts the difference at $1.78 on the day and $39.26 across the 22 days. The free $0.35 credit alone covers about 1.6 million GPT-6.1 Sol input tokens, which is a comfortable way to test the service before topping up.

The pattern in both examples is the same. Caching pays, the named model is the billed model, and the usage records let you check the arithmetic yourself rather than take it on faith. Every call appears in the records with its model, tokens and charge, so a desk day on BestRelay can be audited line by line at the end of it. Nothing about the invoice needs to be taken on trust, which is a modest virtue and a useful one.

Setting up BestRelay takes two settings and a moment of matching the client

Connecting a client to BestRelay takes two settings, the address and a key created under API keys in the dashboard, with the details kept in the documentation. The other half of setup is matching the client to the model, because BestRelay does not pretend every tool speaks every protocol.

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. Tools such as Codex pair with GPT, DeepSeek, Kimi, GLM, Qwen and Grok models instead. Gemini models use only their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible. In the example day, the Claude work runs in Claude Code and the GPT work runs in Codex, while each call is checked against the model it was written for.

Once the pairing is right, the three codes from the morning become easy to place. A 401 means the key was not accepted and can be replaced at any time. A 402 means the prepaid balance needs a Stripe top-up from the published amounts. A 429 means you crossed 120 requests a minute on one key or 8 calls in flight on one account, and the answer is to wait. BestRelay keeps the line-up curated, the billing transparent and the model you name as the model that answers, which is a modest set of promises and rather easier to live with than a surprise on the invoice. Give BestRelay a try on a quiet afternoon, with a small top-up and a model you actually want to use.