BestRelay and Claude Haiku 4.5: pay-as-you-go API relay

BestRelay forwards each request to the model you name

BestRelay is an independent API relay, and its promise is a plain one: it forwards every request to the model provider under the model name the caller wrote. The model you name is the model that answers, and nothing is quietly swapped for a cheaper tier behind your back. For a desk full of small, frequent chores, that rule matters more than any slogan would. You decide which model handles the tidying, the drafting, the labelling, the summarising, and you keep the naming honest.

The line-up is curated and finite: 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 sheet you read is meant to reflect what is actually being served. Prices are quoted per 1,000,000 tokens, which is the unit that makes a month of chores legible.

The twenty names on the sheet today are:

• 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
• GPT-5.6 Sol, GPT-6 Astra, GPT-6.1 Sol, DeepSeek V4 Pro, Kimi K3, GLM-5.3, Qwen3.8 Max, Grok 4.6, Grok 4.7
• Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash, Gemini 4 Argon

Because the model named is the model that answers, choosing Claude Haiku 4.5 for the repetitive half of your day is a decision you can audit later in the usage records, where every call appears with its model, its tokens and its charge. BestRelay does not require you to guess which neighbour of a model actually ran. You write the name; the record shows the name.

A sample desk day, for example, priced in cents

Here is a sample day at your own desk, and everything in it is an example. Suppose you spend the working hours on small, frequent calls: short prompts, tidy answers, the same few tasks repeated across dozens of requests. As an example, take a day of 500,000 input tokens and 100,000 output tokens sent to Claude Haiku 4.5.

On BestRelay, that day costs $0.41. Twenty-two such days cost $9.01. If 350,000 of the input is read from cache instead, the day costs $0.28. The rate card behind those figures is Claude Haiku 4.5 at input $0.4167, output $2.01, cache write $0.4861 and cache read $0.0333 per 1,000,000 tokens.

For the same example at the official list price, the day is $1.00, the 22 days $22.00 and the cached day $0.69. The difference on the day is $0.59; on the 22 days it is $12.99. That gap is not a mystery coupon but arithmetic: the official list price for Claude Haiku 4.5 is input $1, output $5 and cache read $0.1, and BestRelay charges the rate card above it.

• Example day, uncached: $0.41 on BestRelay against $1.00 at the official list price
• Example of 22 such days: $9.01 on BestRelay against $22.00 at the official list price
• Example with 350,000 input tokens read from cache: $0.28 on BestRelay against $0.69 at the official list price

For comparison, as another example, a day of the same shape sent to Claude Sonnet 5.5 costs $1.40 on BestRelay, 22 such days $30.71, and with 350,000 of the input read from cache the day costs $0.94. At the official list price that same day is $2.00, the 22 days $44.00 and the cached day $1.33, a difference of $0.60 on the day and $13.29 on the 22 days. The point of the arithmetic is not that one model fits every chore, but that you can price each chore before you commit to it.

Claude Haiku 4.5 reaches you through Claude Code and Claude Desktop

One practical detail deserves to be stated early. Claude models are served only to the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. If your day is spent in one of those two clients, Claude Haiku 4.5 is available to you through BestRelay without further ceremony.

The same courtesy applies to the other tools. Claude models 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. BestRelay is explicit about these boundaries because a relay that hides them wastes an afternoon.

Connecting a client takes two settings: the address, and a key you create under API keys in the dashboard. The details are in the docs. You will not find either value reprinted here, and you should keep it that way; the address and the key are yours, and the dashboard is where they belong.

Once a client is connected, the day described above can proceed as ordinary work. You write the model name into the request, the request travels under that name, and the usage records later show the same name beside the tokens and the charge. BestRelay does not need to be told twice what you meant, and neither do you.

Prepaid pay-as-you-go balance, billed once per finished call

The money side of BestRelay is deliberately plain. You open with a prepaid pay-as-you-go balance: 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. Nothing is held in advance, so a call in flight is not a call you have already paid for.

Failure is handled with the same directness. Requests the provider rejects are recorded at zero, and when the provider cannot be reached nothing is charged. A failed call therefore costs you nothing and still appears in the records, which is the behaviour you want when a hundred small requests are moving through your afternoon.

• Every new account opens with $0.35 of credit, enough to begin experimenting straight away
• Top-ups run through Stripe checkout at $1.39, $6.94, $13.89, $27.78, $69.44 or $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 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

For example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556. One more figure helps set the scale of the free credit: it alone covers about 1.6 million GPT-6.1 Sol input tokens, so a new account is not a token gesture. It is a working balance.

Limits, records and the shape of a long afternoon

BestRelay sets limits that suit a busy desk without pretending there are none. A single key carries 120 requests a minute, and an account runs 8 calls in flight. Those are ceilings rather than quotas on your ambition; they simply keep the queue orderly while you work through your chores.

Keys are cheap to hold and easy to retire. An account can hold several keys, each one can be voided at any time, and every call appears in the usage records with its model, tokens and charge. If you want to know what a particular afternoon cost, or which model handled which request, the records will tell you without a support ticket.

Privacy is stated plainly as well. The user-written text of requests is kept for 30 days, so abuse can be reviewed and billing questions settled. System prompts and model replies are not stored. BestRelay keeps the material it needs for its stated purposes and lets the rest go.

Put together, the arithmetic is unglamorous in the way you want it to be. A curated line-up of 20 public models, each named as itself; prepaid pay-as-you-go credit with no monthly fee; charges recorded once per finished call; and a sample desk day where Claude Haiku 4.5 costs $0.41 for 500,000 input tokens and 100,000 output tokens. If that shape of pricing suits the way you work, you can open a BestRelay account and start with the $0.35 of credit already sitting there.