BestRelay pricing for Claude Sonnet 5.5, an API relay without fees

BestRelay charges no monthly fee, so a quiet week costs only what you call

The following is an example, plainly labelled as one. You are at your own desk on a Tuesday in an ordinary week. There is no standing subscription line on your card, no monthly fee to forget about, and no minimum spend to justify. You open BestRelay when you need it, name the model you want, and pay for the calls you actually make. On a quiet week, that is a small number of calls and a small balance drawn down.

The mechanics are deliberately plain. BestRelay is an independent API relay, and it runs on a prepaid pay-as-you-go balance. Each call is charged once, after it ends, from the tokens the provider reports, at the rate card prices. Nothing is held in advance. There is no subscription tier to be placed on a plan, and no monthly fee on the account. If you make three calls in a week, you are billed for three calls.

Failures are treated with the same plainness. A request the provider rejects is recorded at zero, and when the provider cannot be reached, nothing is charged. A quiet week, then, can end with a balance that has barely moved, and that is the point of the arrangement rather than a side effect of it.

Starting an account is equally modest. Each new account opens with $0.35 of credit, and top-ups run through Stripe checkout in these 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. So for example, a week in which you barely use the service leaves you with a balance you can top up by a dollar, or ask to have returned within the stated window.

The model you name on BestRelay is the model that answers

BestRelay forwards each request to the model provider under the model name the caller wrote. The model named is the model that answers, and nothing is quietly swapped for a cheaper tier. That single sentence carries most of the argument, because it removes a category of doubt that anyone paying per token will recognise.

The line-up is curated rather than sprawling: 20 public models, each listed under its own name, each with four prices shown for input, output, cache write and cache read. When a provider retires a model id, BestRelay delists it, so the list does not accumulate names that no longer resolve to anything. What remains is a short, checked shelf rather than an open directory of half-attended entries.

The figures on that shelf vary widely, which is another reason to read the rate card rather than assume. As an example, GPT-6.1 Sol is listed at $0.2083 for input and $1.11 for output per 1,000,000 tokens, while Claude Opus 5.5 is listed at $1.74 for input and $8.13 for output. Claude Haiku 4.5 sits at $0.4167 input and $2.01 output, Gemini 2.5 Flash at $0.2778 input and $2.08 output, Grok 4.6 at $0.625 input and $1.81 output, and DeepSeek V4 Pro at $1.53 input and $4.58 output. Gemini 3.1 Pro, Qwen3.8 Max, Kimi K3, GLM-5.3 and the rest of the shelf carry their own four prices in the same way.

BestRelay does not ask you to guess which tier is answering. It asks you to write the name, and then it honours the name, with the price card visible in advance and every call recorded afterwards with its model, tokens and charge.

A sample working week on BestRelay, priced from the rate card

Here is an example week, again labelled as an 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 $0.95. At the official list price, the same day is $3.00, the 22 days are $66.00, and the cached day is $2.06. The difference on the day is $1.60, and on the 22 days it is $35.29.

Now take the same shape of day on GPT-6.1 Sol. As an example, 500,000 input tokens and 100,000 output tokens cost $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 are $44.00, and the cached day is $1.48. The difference on the day is $1.78, and on the 22 days it is $39.26.

Two observations follow, and neither requires arithmetic beyond the figures above. A month of steady use is not a small thing on a rate card, and BestRelay is not the place where that month quietly disappears. The second is that the free credit alone covers about 1.6 million GPT-6.1 Sol input tokens, which is enough to sit with the service before committing anything.

Cache behaviour is worth noticing in both examples, since a desk day often repeats material. Reading input from cache lowers the day’s cost on BestRelay in both cases, and the rate card publishes cache write and cache read prices for every model that has them, so the saving is visible rather than implied.

BestRelay clients and protocols are narrow, and deliberately so

BestRelay is careful about where its models run, and it says so plainly rather than leaving you to find out by trial and error.

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, along with Codex, pair only with GPT, DeepSeek, Kimi, GLM, Qwen and Grok models. If your working day is built around Codex or one of the other editors, the models available to you are named in that list, and the naming is the same on the price card.

Gemini models use only their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible. There is no compatibility shim to configure and no second dialect to learn; the client speaks the protocol the model expects.

Connecting a client takes two settings: the address and a key created under API keys in the dashboard. The details are in the BestRelay docs, and they stay there rather than being reprinted here. This narrowness is a feature of the service as described: a short list of supported clients, a short list of supported models, and no ambiguity about which combination is meant to work.

Usage is bounded too. A key is good for 120 requests a minute, and an account may have 8 calls in flight at once. For a single desk and a single person, those limits leave room for an ordinary working rhythm while keeping the relay’s behaviour predictable.

BestRelay accounts, keys, records and the small sums involved

Your BestRelay account can hold several keys, each of which can be voided at any time. Every call appears in the usage records with its model, tokens and charge, so a quiet week can be checked rather than guessed at. This is the practical companion to a pay-as-you-go balance: no monthly fee to reconcile against, but a record for each call when you want one.

Privacy is handled with restraint. 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. The retention period is stated, its purpose is stated, and the categories excluded are stated as plainly as the ones included.

There is also a way to bring others in. For example, when someone joins through your link, you receive 18% of their top-ups, 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. On an invited member’s top-up of $13.89, the two commissions are $2.50 and $0.5556.

Put together, this is a service shaped for someone who would rather pay for what they use than keep a subscription alive for the sake of it. No monthly fee, no subscription, no minimum spend. A curated shelf of 20 public models at published rates, the model you name being the model that answers, calls billed once after they end, rejected calls recorded at zero, and every charge visible in the records. BestRelay is the kind of relay you set up once at your desk and then let a quiet week cost exactly what it was worth.