Cursor with BestRelay: an API relay for GPT-6.1 Sol

BestRelay in Cursor: an example day when the model you name is the one that replies

Here is an example day at your own desk, and it is an example rather than a record of anyone's actual usage. You open Cursor in the morning with a BestRelay key already in place, and you start with GPT-6.1 Sol, because that is the model you have decided to work with. BestRelay is an independent API relay, and it forwards each request to the model provider under the model name the caller wrote, so the model named is the model that answers and nothing is quietly swapped for a cheaper tier. That single property is the whole point of the arrangement, and it is worth stating plainly before anything else happens.

Two settings get you there: the address and a key created under API keys in the dashboard. The details are in the docs, and this article prints neither of them. You paste the key into Cursor, pick GPT-6.1 Sol from the list, and ask for a first pass at a small refactor.

By midday the work has drifted, as work does. You move to DeepSeek V4 Pro for the plumbing and Kimi K3 for a summary of the diff, then to Grok 4.7 for a quick check of a generated snippet. Every one of these is a model in the BestRelay line-up, listed under its own name, and every one is available to tools such as Cursor, Cherry Studio, Cline, OpenCode, Codex, any SDK or cURL.

The arithmetic stays plain. GPT-6.1 Sol is $0.2083 per 1,000,000 input tokens and $1.11 per 1,000,000 output tokens on BestRelay. Grok 4.7 is $0.625 and $1.94. Kimi K3 is $1.39 and $7.01. DeepSeek V4 Pro is $1.53 and $4.58. 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 quiet afternoon costs what it costs.

The BestRelay line-up is curated, named plainly, and priced in the open

The line-up holds 20 public models, and the curation is deliberate. Each one is listed under its own name with four prices: input, output, cache write and cache read. You are not reading a vague description of a tier; you are reading a number attached to a name you can type. When a provider retires a model id, BestRelay delists it, so the sheet of models you are reading is not padded with names that no longer resolve.

Among the entries that pair with Cursor and the other OpenAI-compatible clients are GPT-5.6 Sol, priced at $0.4167 for input and $2.29 for output per 1,000,000 tokens, GPT-6 Astra at $0.7639 and $3.96, GPT-6.1 Sol at $0.2083 and $1.11, Qwen3.8 Max at $0.8333 and $2.57, GLM-5.3 at $0.625 and $1.94, and Grok 4.6 at $0.625 and $1.81. Cache reads are priced too, and they are small: $0.0333 for GPT-5.6 Sol, $0.0431 for GPT-6.1 Sol, $0.1014 for Qwen3.8 Max, and $0.1389 for GLM-5.3 and for both Grok models.

Two families sit outside that OpenAI-compatible set, and BestRelay is straightforward about it. Claude models are served only to the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. Gemini models use only their native protocol, through Gemini CLI or direct generateContent calls, and are not OpenAI-compatible. Their rates still appear on the same card: Gemini 3.1 Pro is $1.74 input and $9.93 output per 1,000,000 tokens, Gemini 3.8 Flash is $1.25 and $5.56, Gemini 4 Argon is $1.67 and $8.06, and Gemini 2.5 Flash is $0.2778 and $2.08.

Curated does not mean vague. It means the list is short, the names are exact, and the prices beside them are the prices you will be charged.

Keys, limits and records: what the dashboard keeps about your Cursor sessions

An account can hold several keys, and each one can be voided at any time. That matters for a desk-bound workflow: you can keep a separate key for the work you do in Cursor and another for a script, and retire the one you no longer trust without disturbing the rest. Keys are created under API keys in the dashboard, and a client needs the address and the key to connect. The docs hold those details, which is where they belong.

There are two limits worth knowing before you queue a batch of work. One key may send 120 requests a minute, and one account may have 8 calls in flight at a time. Both are ceilings rather than targets, but they shape how you batch: a wide sweep of prompts is better spread across a few minutes than fired in a single burst.

Every call appears in the usage records with its model, the tokens it used and the charge it produced. Because the record carries the model name as well as the tokens, you can check afterwards that the reply you received came from the model you asked for. This is where the promise at the centre of BestRelay becomes checkable rather than merely stated: name the model, then look at the record and see that name sitting there beside the charge.

Tools that speak the OpenAI-compatible protocol, including Cursor, Cherry Studio, Cline, OpenCode, Codex, any SDK or cURL, pair with GPT, DeepSeek, Kimi, GLM, Qwen and Grok models. Knowing which family belongs where saves you a wasted call and a confused afternoon, and the usage records will show you the answer either way.

Pay-as-you-go billing, prepaid and settled after each call

BestRelay runs on a prepaid pay-as-you-go balance. There is no monthly fee, no subscription and no minimum spend, so a light week costs a light amount and a heavy week costs what it actually used. Each call is charged once, after it ends, from the tokens the provider reports, at the rate card prices. Nothing is held in advance, which means the balance in your account is the balance you have actually spent.

Failure is handled without drama. Requests the provider rejects are recorded at zero, and when the provider cannot be reached nothing is charged. A rejected prompt therefore costs nothing to try, and an unreachable provider does not quietly drain a balance you were saving for the next real request.

Each new account opens with $0.35 of credit, enough to make a handful of small calls before you top up. Top-ups run through Stripe checkout in these amounts: $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. The smaller amounts suit a day of experimenting; the larger ones suit a stretch of steady use, and either way you are choosing the size rather than being sold a package.

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. What BestRelay offers is a curated route to the models, priced from its own card and settled per call, with the model named by the caller carried through to the provider.

Worked examples: what a day of Cursor traffic costs through BestRelay

For example, take Claude Sonnet 5.5 through the clients that serve it. 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, so the difference is $0.60 on the day and $13.29 across the 22 days.

For example, take GPT-6.1 Sol, the model you met at the start of this day. 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, a difference of $1.78 on the day and $39.26 across the 22 days. The free credit alone covers about 1.6 million GPT-6.1 Sol input tokens.

There is a referral scheme as well. 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. For example, on an invited member's top-up of $13.89, the two commissions are $2.50 and $0.5556.

One last practical note on privacy: 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. With the model named, the rate card published, the charge settled per call and the record kept in view, BestRelay leaves you with a desk, a key and a model you chose yourself. Try BestRelay with the model you already had in mind.