Quanta pricing for Claude Sonnet 5: $6.14 per 1M output tokens
01 · What does it mean to run Claude Sonnet 5 on Quanta?
Quanta is an API relay built for teams with heavy, sustained workloads. Instead of hiding capacity behind vague tiers, Quanta publishes its concurrency and rate limits up front and meters every call on real tokens. For Claude Sonnet 5 (claude-sonnet-5), Quanta lists input at $1.22 per 1M tokens and output at $6.14 per 1M tokens, with cache writes at $1.54 and cache reads at $0.1069 per 1M tokens. Those four numbers are separated on purpose, so a workload that leans on cached context is priced differently from one that does not.
Quanta runs on a prepaid balance. There is no monthly fee and no minimum spend, which makes the model easy to trial: signing up adds a $0.50 trial credit, and each call is metered individually. The dashboard lists the tokens and the charge of every call, so input, output, cache writes and cache reads are visible rather than bundled into a mystery line item. For a team that processes documents, code or long conversations, that separation matters because cache reads at $0.1069 per 1M tokens can be a fraction of the input price, and the dashboard lets you see exactly when that happens.
The base address for the API is https://api.quantarelay.com, and one key plus one balance covers the OpenAI-compatible (/v1), Anthropic native and Gemini native protocols. You do not juggle separate accounts per protocol; you fund one balance on Quanta and route whichever supported model your client needs.
02 · Which clients can I use with Claude Sonnet 5?
Claude models on Quanta work only from the official Claude Code and Claude Desktop clients, and you need Claude Code 2.1.285 or newer. That pairing is deliberate: one key and one balance on Quanta cover multiple protocols, but Claude Sonnet 5 itself is reached through Claude Code or Claude Desktop, not through other editors or raw HTTP calls.
If your team also runs GPT, DeepSeek or the other non-Claude, non-Gemini models, those can be paired through clients such as Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL. Gemini models use the native protocol only, through Gemini CLI. Keeping the pairings explicit means the client you already use decides which protocol your single Quanta key speaks, and Quanta keeps the billing unified across all of them.
03 · How do I fund a Quanta account, and what does a top-up look like?
Quanta is pay-as-you-go only. You choose a top-up amount from the published list: $1.39, $6.94, $13.89, $27.78, $69.44 or $138.89. Charges are made in CNY, and the dollar figures are approximate USD references. Because billing is prepaid, you only spend what you loaded, and unused balance can be refunded within 15 days of the top-up.
As an example, and this is only an illustration: say your team runs a sustained batch job that sends 10 million input tokens to Claude Sonnet 5 in a month with no cache hits. At the listed $1.22 per 1M input tokens, that input side alone would come to about $12.20, before any output tokens. The point of the example is to show the arithmetic is transparent on Quanta: tokens in, tokens out, each priced separately, all visible per call in the dashboard.
If you bring others along, Quanta pays referral rewards of 19% of a referred account's top-ups, and 5% one level further. Those rewards can be withdrawn from $27.78 or moved into the balance, so a team that grows its usage can recycle part of that spend back into future calls.
04 · What limits and concurrency does Quanta publish?
Quanta spells out the envelope instead of burying it. Per key, the limit is 120 requests per minute; per account, 8 requests in flight; each request body is capped at 25 MB; and an account can hold 20 live keys. Streams have no total-time cap and only end after 300 seconds of upstream silence, which suits long, sustained generation jobs.
These limits are enforced, not aspirational, so a heavy workload can plan around them. If your team needs room to grow, the practical levers are the number of keys and the in-flight allowance, all managed from the same dashboard that shows per-call tokens and charges. Because Quanta is built for concurrency, you can spread load across several keys within the published 120 requests per minute per key ceiling and keep 8 requests in flight at once without guessing at hidden caps.
05 · How do company payments and invoices work on Quanta?
For company spend, Quanta accepts bank transfer from a company account. Invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. There is no separate support email to remember; the dashboard messages and the Contact page are the route to the team.
That matters for finance teams because the paper trail matches the usage: every call is metered on real tokens, the dashboard lists tokens and charge per call, and the invoice conversation happens in the same place where the spend is recorded. A team that needs an advance quote before committing budget can request one through those same channels, and the prepaid model means the quote and the eventual invoice line up with what the dashboard already shows.
06 · What happens to my prompts, and how do I get started?
Quanta keeps the text users send for 30 days for abuse handling and billing disputes, then deletes it; model answers are not stored. So the retention window is short and specific, and it exists for dispute resolution rather than for building profiles.
Getting started is short: open an account on Quanta, take the $0.50 trial credit, create a key, and point Claude Code 2.1.285 or newer, or Claude Desktop, at Quanta for Claude Sonnet 5. Pick a top-up that matches your expected volume, watch the first calls land in the dashboard with their token counts and charges, and scale from there. Quanta is built for teams that want concurrency, rate limits and invoices spelled out, and Claude Sonnet 5 at $1.22 per 1M input tokens is a straightforward place to begin.