Claude Opus 5 API Relay Pricing on Quanta for Heavy Traffic

01 · What Claude Opus 5 costs per 1M tokens on Quanta

Quanta is built for teams running heavy, sustained workloads, and Claude Opus 5 sits at a price point that fits that kind of traffic. Every call is metered on real tokens, and the four token types are priced separately, so the number you see on the dashboard is the number you can reconcile against your own usage. Quanta lists Claude Opus 5 (claude-opus-5) at $1.97 per 1M input tokens, $9.83 per 1M output tokens, $2.42 per 1M cache write tokens and $0.1222 per 1M cache read tokens.

Those figures are approximate USD amounts; charges are made in CNY. Nothing sits behind a subscription. Quanta uses a prepaid balance with no monthly fee and no minimum spend, so a team can start with a small top-up, measure real consumption, and add funds only when the workload justifies it.

Here is an example, not a quotation: if a team sends 10 million input tokens and receives 3 million output tokens in a month with no cache hits, the input side is 10 million times $1.97, or $19.70, and the output side is 3 million times $9.83, or $29.49. That example comes to roughly $49.19 for the month. Because Quanta lists the tokens and the charge for every individual call, you can check where the total came from rather than taking it on trust.

02 · Setting up Claude Opus 5 in Claude Code and Claude Desktop

Claude Opus 5 works from the official Claude Code and Claude Desktop clients. If your team already writes in Claude Code, you keep that workflow and point it at Quanta instead of changing how you work. Claude Code 2.1.285 or newer is required for Claude Code pairing.

The setup path is deliberately short:

• › Open Quanta and create an account. Sign-up includes $0.50 in trial credit, which is enough to confirm that a request is moving end to end before a larger top-up.
• › Add funds to the prepaid balance. Available top-up amounts are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89.
• › Create a live key in the dashboard and use it in the official Claude Code or Claude Desktop client for Claude Opus 5.
• › Run a first request, then open the dashboard and read the token counts and charge recorded against that call.

Quanta speaks the OpenAI-compatible protocol, the Anthropic native protocol and the Gemini native protocol, and one key and one balance cover all three. For Claude models, use the Anthropic native path through the official Claude Code and Claude Desktop clients. Claude models are not paired with Cursor, Cherry Studio, Cline, OpenCode, Codex, an SDK or cURL; those clients pair with GPT, DeepSeek and the other non-Claude, non-Gemini models.

03 · Rate limits and concurrency for sustained team workloads

Heavy workloads are rarely a single stream of requests. They are batches, retries, long-running sessions and several engineers working at once. Quanta publishes the limits it enforces, so you can design around them instead of guessing:

• › 120 requests per minute per key.
• › 8 requests in flight per account.
• › 25 MB per request body.
• › 20 live keys per account.

The concurrency ceiling is the one to think about first: eight requests in flight per account is a firm, account-level number, and more keys do not raise it. That makes it straightforward to schedule batch jobs, keep interactive work responsive and avoid piling work onto a single session. Rate limits are per key, so distributing traffic across up to 20 live keys gives you room to separate services, environments or teammates without colliding at 120 requests per minute.

Long responses are handled too. Streams carry no total-time cap and end only after 300 seconds of upstream silence, which suits extended generation rather than cutting it off on a fixed timer. Because the account-level in-flight count is published and enforced, you can size a workload before it reaches Quanta, and you can check the recorded charge for each call afterward to see whether your batching strategy is paying off.

04 · Prepaid billing, invoices and refunds on Quanta

Quanta charges as you go, on metered tokens. There is no monthly fee, no minimum spend and no pass or bundle. The dashboard lists the tokens and the charge of every call, which matters when several teams share one account and someone needs to explain a line item.

For companies, 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; both routes sit inside the service you already use.

Refunds are straightforward as well: unused balance can be refunded within 15 days of the top-up. That short window is worth noting at the start, so top up against expected consumption rather than against a vague forecast.

Referrals work in the same transparent way. A referred account's top-ups generate 19% back, and 5% applies one level further. Rewards can be moved into the balance or withdrawn once they reach $27.78. For a team that is already funding a heavy Claude Opus 5 workload, that is a direct credit against the next top-up rather than a promise.

05 · Running Claude Opus 5 alongside other models on one balance

Most teams end up with more than one model in the stack, and Quanta keeps them on the same balance and the same key. Claude Opus 5.5 (claude-opus-5-5) is listed at $1.79 input and $8.46 output per 1M tokens, with cache writes at $2.11 and cache reads at $0.0819. Claude Sonnet 5.5 (claude-sonnet-5-5) is $1.25 input, $6.28 output, $1.57 cache write and $0.0917 cache read. Claude Haiku 4.5 (claude-haiku-4-5-20251001) sits at $0.4167 input, $1.97 output, $0.4861 cache write and $0.0333 cache read.

Worked examples make the shape clearer. At a 3:1 input-to-output ratio with no cache hits, Claude Opus 5.5 runs about $34.58 for 10 million tokens a month and about $345.83 for 100 million. Claude Sonnet 5.5 comes to roughly $25.07 and $250.69 on the same basis. For comparison within the same balance, GPT-6.1 Sol is about $4.10 and $40.97, and Gemini 2.5 Flash about $6.49 and $64.93. These are examples, not quotes, and each one is built from the published per-token rates.

Protocol pairing stays clean. Gemini models use the native protocol only, through Gemini CLI. GPT, DeepSeek and the other non-Claude, non-Gemini models work with Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL. Claude Opus 5 remains with Claude Code and Claude Desktop.

06 · Data handling, invoicing and your first week on Quanta

Prompt handling on Quanta is stated plainly. The text you send is kept for 30 days for abuse handling and billing disputes, then deleted. Model answers are not stored. That distinction is useful for teams who want a clear answer about what persists and for how long.

A sensible first week looks like this. Day one: create the account, claim the $0.50 trial credit, top up at $1.39 and run a small Claude Opus 5 request from Claude Code or Claude Desktop, then read the token breakdown on the call. From there, size the top-up against the request pattern you actually saw, and use the published limits of 120 requests per minute per key and 8 requests in flight per account to decide how many keys and how much concurrency your workload needs. Keep invoices and advance quotes moving through the Contact page or the Messages page of the dashboard when finance needs documentation.

That is the whole Quanta proposition in practice: pay-as-you-go metering on real tokens, limits that are published and enforced, one balance across protocols, and a dashboard that shows what each call cost. Open an account, add the balance your workload needs, and put Claude Opus 5 to work through the official clients.