Cursor Pay-as-you-Go API Pricing: Quanta for a 5-Person Startup

01 · Why a Five-Person Startup Should Plan Its AI API Costs in the Open

Shipping AI features with a small team rarely fails because of a lack of ambition. It fails because the underlying costs, limits and billing rules are hard to predict. A feature that behaves well during development can behave very differently when real users start sending long prompts, running parallel requests and streaming long responses.

Quanta is an API relay built for teams and heavy, sustained workloads. The point is straightforward: concurrency, rate limits and invoices are spelled out instead of hidden, so a five-person startup can reason about capacity and cost before a release goes out. Quanta uses a prepaid balance, charges per call on real tokens, and lists the tokens and charge of every call in the dashboard. There is no monthly fee and no minimum spend, so a team can start small, measure what it actually uses, and top up when the numbers justify it.

For a team that wants to move quickly without signing a long contract or guessing at overhead, that model is easier to manage. You buy what you consume. Sign-up includes $0.50 in trial credit, which is enough to make a first call and see how the dashboard records it before committing meaningful budget.

02 · Choosing Models Without Locking Your Team Into One Client

Quanta supports OpenAI-compatible (/v1), Anthropic native and Gemini native protocols, and one key and one balance work across all three. That matters for a startup because different clients have different pairing rules, and Quanta is precise about them.

Cursor, Cherry Studio, Cline, OpenCode, Codex, OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models available on Quanta. The catalogue includes GPT-5.6 Sol, GPT-6 Astra, GPT-6.1 Sol, 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. If your engineers work in Cursor, Quanta gives them a single key and a single prepaid balance to work with across those models.

Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. The Claude family on Quanta includes 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. Gemini models use the native protocol only, through Gemini CLI, and the catalogue covers Gemini 2.5 Flash, Gemini 3.1 Pro, Gemini 3.8 Flash and Gemini 4 Argon.

• › Cursor and similar clients: GPT, DeepSeek, Qwen, Kimi, Hunyuan, Grok and GLM models.
• › Claude Code and Claude Desktop: Claude models.
• › Gemini CLI: Gemini models through the native protocol.

One account, one balance, one dashboard — and a clear answer to which client belongs with which model.

03 · What Runaway Token Costs Actually Look Like on Quanta

Quanta meters each call on real tokens. Input, output, cache writes and cache reads are priced separately, so a team can see where the spend comes from rather than receiving a single opaque figure at the end of a billing period.

Prices are listed per 1M tokens in approximate USD, and charges are made in CNY. For example, Claude Opus 5.5 is listed at $1.79 for input, $8.46 for output, $2.11 for cache write and $0.0819 for cache read. Claude Sonnet 5.5 is listed at $1.25, $6.28, $1.57 and $0.1097. On the lighter end, Qwen3.7 Plus is listed at $0.1667 for input, $0.4861 for output and $0.0486 for cache read, while GPT-6.1 Sol is listed at $0.2083 for input, $1.01 for output and $0.0403 for cache read. Gemini 2.5 Flash is listed at $0.2361 for input, $1.89 for output, $0.2361 for cache write and $0.0236 for cache read.

To make the arithmetic concrete, these are examples at a 3:1 input-to-output ratio with no cache hits. A team using 10 million tokens a month on Claude Opus 5.5 would spend approximately $34.58; at 100 million tokens, approximately $345.83. The same examples for Claude Sonnet 5.5 are approximately $25.07 and $250.69. For GPT-6.1 Sol, the examples are approximately $4.10 and $40.97, and for Gemini 2.5 Flash, approximately $6.49 and $64.93.

Those are examples, not quotes for your product. The point is that you can build a rough budget before launch and then check it against the per-call records in the dashboard.

04 · Concurrency and Rate Limits Your Team Can Actually Plan Around

Published limits are useful only if they are enforced consistently and stated plainly. Quanta publishes and enforces 120 requests per minute per key, 8 requests in flight per account, 25 MB per request body and 20 live keys per account.

For a five-person startup, that shape is workable and legible. Eight requests in flight per account means parallel work is bounded, so a batch job cannot quietly consume the entire capacity your product needs. Twenty live keys allow separation between production, staging and individual experiments without creating an unbounded number of credentials. The 25 MB request body limit is large enough for substantial payloads while remaining a defined boundary rather than a surprise.

Streams have no total-time cap and only end after 300 seconds of upstream silence. In other words, a long-running stream is not cut off by an arbitrary duration limit; it ends when the upstream stops sending for 300 seconds. For teams shipping streaming features, that is a meaningful difference between a documented boundary and a vague promise.

Because the limits are published, you can design your retry logic, queue depth and concurrency settings around them instead of discovering them during an incident.

05 · Prepaid Balance, Company Invoices and a Clean Paper Trail

Quanta bills against a prepaid balance, and the dashboard lists the tokens and charge of every call. That level of detail is what makes a shared team account workable: each person can see what was spent, on which model, and for which request.

Top-ups are available at $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. A small team can begin with a modest amount and increase it as usage becomes predictable. Unused balance can be refunded within 15 days of the top-up, which keeps early experiments low-risk.

For formal purchasing, bank transfer from a company account is supported, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. There is no support email address to hunt for; the same two routes cover both questions and paperwork.

There is also a referral programme: 19% of a referred account's top-ups, and 5% one level further. Rewards can be withdrawn from $27.78 or moved into the balance.

06 · Getting Your Five-Person Team Started on Quanta

Start with the clients your engineers already use. If the team works in Cursor, pick from the GPT, DeepSeek, Qwen, Kimi, Hunyuan, Grok and GLM models on Quanta and point the client at the OpenAI-compatible protocol. If a developer is working in Claude Code or Claude Desktop, use Claude models there, and reach Gemini models through Gemini CLI with the native protocol.

Then create keys deliberately, keep the count inside the 20 live keys limit, and set your client-side concurrency so that combined traffic stays within 8 requests in flight per account. As soon as traffic arrives, read the dashboard: every call shows its tokens and its charge, broken down across input, output, cache writes and cache reads.

From there, top up only what the measured usage supports. The prepaid balance, the absence of a monthly fee and the absence of a minimum spend mean a five-person startup can start with the $0.50 trial credit, expand gradually, and move to company bank transfer and invoices when the workload justifies it.

Quanta does not hide the mechanics. It publishes the prices, the limits, the concurrency ceilings and the billing rules, then lets the team decide what to build. For a startup shipping AI features with a handful of engineers, that clarity is the practical advantage.