Setting up BestRelay with Claude Code as an API proxy

BestRelay sends the review to the model you actually named

Here is an example day at your own desk, set out plainly as an example. You maintain a small open-source library, and a pull request has been sitting overnight, waiting for a careful second reading. You open Claude Code, point it at the repository, and ask for a review that checks the public API for accidental breakage. You decide the change deserves a strong model, so you name Claude Opus 5.5. BestRelay forwards that request to the provider under the name you wrote, and Claude Opus 5.5 is the model that answers. Nothing is quietly swapped for a cheaper tier on the way through.

That single sentence is the whole idea. BestRelay is an independent API relay, and its line-up is curated rather than scraped: 20 public models, each listed under its own name with four prices for input, output, cache write and cache read. When a provider retires a model id, BestRelay delists it, so the names in front of you are names you can actually call. For your review, Claude Opus 5.5 is priced at $1.74 per 1,000,000 input tokens, $8.13 per 1,000,000 output tokens, $2.01 for cache write and $0.0778 for cache read. The official list price for the same model is input $4, output $20, cache read $0.2. You glance at the two columns, decide which suits your budget, and keep working.

For a maintainer, that predictability matters more than a long menu of half-documented options. You know which model you asked for, you know its rate, and you know what the answer will draw from the balance.

The rate card turns a day of review into a number you can check

For example, take a day of 500,000 input tokens and 100,000 output tokens on Claude Sonnet 5.5. On BestRelay that day costs $1.40, and 22 such days cost $30.71. If 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 is $0.60 on the day and $13.29 across the 22 days. These are examples, not quotas.

The same arithmetic reads differently for GPT-6.1 Sol, which is a sensible companion when you want a quicker pass over formatting and lint noise. As an example, a day of 500,000 input tokens 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. The difference on the day is $1.78, and on the 22 days $39.26.

Other corners of the line-up are priced in the same way, in the same four columns. DeepSeek V4 Pro lists $1.53 input, $4.58 output and $0.0514 cache read per 1,000,000 tokens; Gemini 2.5 Flash lists $0.2778 input, $2.08 output, $0.2778 cache write and $0.0264 cache read. For a maintainer who reads a hundred diffs a month, the practical point is simple: the model you name is the model that answers, and the rate card tells you what the answer will cost before you ask for it.

A prepaid balance is charged once, after each call ends

Billing on BestRelay is prepaid pay-as-you-go. There is no monthly fee, no subscription and no minimum spend; you put money in and spend it as you go. Each call is charged once, after it ends, from the token counts the provider reports, at the rate card prices for the model that ran. Nothing is held in advance, so an idle afternoon with no requests costs nothing at all.

The failure cases are handled with the same plainness. Requests the provider rejects are recorded at zero, and when the provider cannot be reached, nothing is charged. A relay that quietly billed for work it never delivered would be a poor companion for a small project budget, and BestRelay does not do that.

Every new account opens with $0.35 of credit, which is enough to see the service behave properly before you commit anything. As an example, that free credit alone covers about 1.6 million GPT-6.1 Sol input tokens, enough for a handful of small reviews while you settle on a habit. Top-ups run through Stripe checkout in set amounts: $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. The smaller amounts suit a weekend project; the larger ones suit a maintainer with a busy merge queue. 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, which is worth knowing before you treat a referral payment as spending money.

Claude models arrive through Claude Code and Claude Desktop

Client support is specific rather than generous-sounding. Claude models are served to Claude Code and Claude Desktop, version 2.1.285 or newer. They do not work in Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL, and no amount of fiddling will change that. Those tools, along with Codex, pair with GPT, DeepSeek, Kimi, GLM, Qwen and Grok models, so if your workflow runs through Codex, the model you name should come from that family.

Gemini models follow their own path: their native protocol, through Gemini CLI or direct generateContent calls. They are not OpenAI-compatible, and BestRelay does not pretend otherwise. The line-up contains Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash and Gemini 4 Argon, each with its own prices, and each reachable in the way its own tooling expects.

Connecting a client takes two settings: the address, and a key created under API keys in the dashboard. The documentation explains both; the address and any setting are deliberately not printed here, because a public relay address pasted into a blog post is a public relay address. Limits are stated plainly as well: 120 requests a minute on one key, and 8 calls in flight on one account. For a desk where one person is reading diffs, that is room to keep several reviews moving without turning your account into a stress test.

Usage records, privacy and referrals stay visible to you

An account can hold several keys, and each key can be voided at any time, which is a reasonable arrangement when a laptop changes hands or a colleague needs a separate trail. Every call appears in the usage records with its model, its tokens and its charge, so at the end of the month you can reconstruct what you spent on what. The same records make the earlier examples checkable: you can compare a $1.40 day against a $0.94 cached day and see exactly where the difference came from.

Privacy is described in plain terms. The user-written text of your requests is kept for 30 days, so abuse can be reviewed and billing questions can be settled. System prompts and model replies are not stored. For a maintainer whose prompts may contain unreleased code or private issue text, that distinction is the one to hold on to.

BestRelay also pays for referrals, which is a modest way to share a tool you rely on. 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. Gift credit and commission cannot be refunded, so treat that money as credit rather than cash. That is the example day end to end: a named model, a published rate, a prepaid balance charged once per call, and a record you can read afterwards. If it suits you, BestRelay is ready when you want to try it.