Cursor API Relay Setup: Quanta Pricing at 120 Requests/Minute

01 · What Cursor Can Run Through Quanta

Cursor is one of the clients Quanta supports directly, because Cursor connects to an OpenAI-compatible provider through a base address and a key. Quanta publishes that OpenAI-compatible route as /v1, and the same key and the same balance also cover Quanta's Anthropic native and Gemini native protocols. For a Cursor team, that means the models available through the OpenAI-compatible route are the practical choice: GPT-6.1 Sol, GPT-5.6 Sol, GPT-6 Astra, DeepSeek V4 Flash, DeepSeek V4 Pro, Qwen3.7 Plus, Qwen3.8 Max, Kimi K3, Hunyuan 4, Grok 4.6, Grok 4.7 and GLM-5.3.

Two boundaries are worth stating plainly, because getting them wrong wastes a team's afternoon. Claude models are not available through Cursor. Quanta serves Claude Fable 5, Claude Fable 5.1, Claude Haiku 4.5, Claude Opus 5, Claude Opus 5.5, Claude Sonnet 5 and Claude Sonnet 5.5 exclusively through the official Claude Code and Claude Desktop clients, using Claude Code 2.1.285 or newer. Gemini models are also separate: Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash and Gemini 4 Argon use the native protocol only, through Gemini CLI, and are never described as OpenAI-compatible.

For teams whose daily work happens inside Cursor, Quanta's OpenAI-compatible route keeps the workflow familiar while putting the operational details on the page instead of behind a sales conversation.

02 · Setting Up the Cursor Connection on Quanta

The setup is short because Quanta keeps the moving parts to a base address, a key and a model name. First, create an account on Quanta and sign in to the dashboard. Your dashboard is where you create keys, and it is where the API base address for your account is shown. Enter that base address in Cursor's OpenAI-compatible provider settings, paste in the key you created, and select one of the supported models by its exact identifier, such as gpt-6.1-sol, deepseek-v4-pro or qwen3.7-plus.

• › Use the base address exactly as Quanta displays it in the dashboard, rather than reconstructing it by hand.
• › Create one key per workload or per environment so that usage stays attributable and a single key can be replaced without disturbing the rest.
• › Start with a small request to confirm the pairing, then run a representative prompt set to see how the model behaves on your real code and documents.
• › Check the dashboard after the first calls. Quanta lists the tokens and the charge of every call, so you can confirm that the request you sent is the request that was metered.

Because Quanta meters on real tokens rather than on a seat count or a bundle, the first thing a team sees is a per-call record. Input tokens, output tokens, cache writes and cache reads are each priced separately, and each of those lines is visible in the dashboard. There is no monthly fee and no minimum spend, so a small trial costs what the trial actually uses.

03 · Rate Limits and Concurrency for Sustained Runs

Cursor sessions are rarely a single request. A team may have several agents editing files, a background indexer running, and a colleague reviewing a diff, all at once. Quanta publishes and enforces the limits that apply to that pattern, so capacity planning does not depend on guesswork.

• › 120 requests per minute per key. Each key carries its own request allowance.
• › 8 requests in flight per account. This is the account-level concurrency ceiling for work running at the same time.
• › 25 MB per request body. Larger payloads need to be split before they are sent.
• › 20 live keys per account. Enough separation for multiple services, environments and contributors without sprawl.
• › Streams have no total-time cap and end only after 300 seconds of upstream silence, so long-running generations are not cut off on a timer.

Those numbers are the practical answer to the question teams usually ask about a relay service: what happens when the load is heavy and sustained? On Quanta, the answer is written down. If a workflow needs more parallel in-flight work than eight requests, the adjustment is to distribute it across keys and accounts deliberately, not to discover a ceiling in production. Quanta also keeps prompt text for 30 days for abuse handling and billing disputes and then deletes it; model answers are not stored.

04 · Pay-as-You-Go Pricing on Quanta

Quanta is pay-as-you-go only. Every call is metered on real tokens, with input, output, cache writes and cache reads priced separately, and the dashboard shows the tokens and the charge for each call. Prices per 1M tokens are listed in approximate USD figures, and charges are made in CNY.

For Cursor workloads, several of the OpenAI-compatible models sit at accessible rates. GPT-6.1 Sol is $0.2083 per 1M input tokens, $1.01 per 1M output tokens and $0.0403 per 1M cache read tokens. Qwen3.7 Plus is $0.1667 input, $0.4861 output and $0.0486 cache read. DeepSeek V4 Flash is $0.2917 input, $1.18 output and $0.006944 cache read. GPT-6 Astra is $0.8333 input, $4.19 output and $0.0833 cache read. Models with cache writes also price that path separately, so a repeated-context workflow can be costed before it runs.

As an example, GPT-6.1 Sol at a 3:1 input-to-output ratio with no cache hits comes to roughly $4.10 for 10 million tokens a month and roughly $40.97 for 100 million. These are examples of monthly volumes, not quotes. Teams can also earn referral rewards of 19% of a referred account's top-ups, plus 5% one level further, withdrawable from $27.78 or moved into the balance.

The balance is prepaid. Top-ups are available at $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89, and unused balance can be refunded within 15 days of the top-up. New accounts receive $0.50 in trial credit on sign-up, which is enough to put the first real request through the pipeline and read the first line of the usage record.

05 · Invoices, Billing and Opening a Quanta Account

Teams buying for a company need a purchase trail, not just a card receipt. Quanta supports bank transfer from a company account, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. The same prepaid balance and the same per-call record apply, so a finance reviewer can trace a line item back to the tokens behind it.

That combination is the reason to consider Quanta for Cursor in the first place. Concurrency is published, rate limits are enforced against stated numbers, prices are per token with each cache path broken out, and every call is itemised in the dashboard. There is no monthly fee, no minimum spend and no plan to buy before the work arrives.

Open an account on Quanta, add the base address and a key to Cursor, and run a first request against gpt-6.1-sol or deepseek-v4-pro. Then read the dashboard line by line: the tokens, the charge, the model, the key. If the numbers hold up under your own workload, that is the whole evaluation.