Kimi K3 pricing: one relay service, one balance, 120 RPM at Quanta

01 · Which protocol carries a Kimi K3 request?

Kimi K3 sits in the broad, non-Claude, non-Gemini group of models on Quanta, so it travels over the OpenAI-compatible /v1 path. That single detail decides which clients you can pair with it: Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs and cURL all pair with Kimi K3 and the other GPT, DeepSeek, Qwen, Hunyuan, Grok and GLM models on Quanta. One key and one balance cover all of it.

Quanta exposes three wire protocols, and the differences are specific rather than decorative. The OpenAI-compatible /v1 endpoint handles the OpenAI SDKs and the third-party clients listed above. Anthropic native serves the Claude family, and Gemini native serves the Gemini family. Gemini models use the native protocol only, through Gemini CLI; they are not OpenAI-compatible. Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. These are exact pairings, stated plainly so a team can pick a client and a model in the same breath.

All three protocols point at the same base address, api.quantarelay.com, and all three draw down the same prepaid balance. There is no need to hold a separate account per protocol, and no need to reconcile several bills at month end.

02 · How does one balance cover three protocols?

Quanta is built for teams and heavy, sustained workloads, so the accounting is designed to be read rather than inferred. There is a prepaid balance, no monthly fee and no minimum spend. Every call is metered on real tokens, and input, output, cache writes and cache reads are priced separately. The dashboard lists the tokens and the charge of every call, which means a line item can be traced back to the request that produced it.

• › Start small. Sign-up includes $0.50 in trial credit, enough to place real calls before a team commits a larger top-up.
• › Top up by amount. The available top-ups are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. Unused balance can be refunded within 15 days of the top-up.
• › Bring colleagues. Referral rewards are 19% of a referred account's top-ups and 5% one level further. Rewards can be moved into the balance or withdrawn from $27.78.
• › Pay as you go. There are no passes, bundles or recurring commitments. Spend tracks usage, and the balance shows what is left.

Quote prices are approximate USD figures, and charges are made in CNY. That is stated up front on Quanta rather than buried in a footnote, and it is one of the reasons teams can budget against the balance without guessing.

03 · What does a request cost on Quanta?

Quanta lists prices per 1M tokens, split across input, output, cache write and cache read. For Kimi K3, the input price is $1.31 and the output price is $6.54, with cache reads at $0.1306 and no cache write price listed. Around it, the range is wide enough to serve different jobs: GPT-6.1 Sol is $0.2083 on input and $1.01 on output, Qwen3.7 Plus is $0.1667 and $0.4861, Gemini 2.5 Flash is $0.2361 and $1.89, and Claude Opus 5.5 is $1.79 and $8.46.

Illustrative monthly totals help put those unit prices in context. These are examples, calculated at a 3:1 input-to-output ratio with no cache hits:

• › Claude Opus 5.5: 10 million tokens a month ≈ $34.58; 100 million ≈ $345.83.
• › Claude Sonnet 5.5: 10 million tokens a month ≈ $25.07; 100 million ≈ $250.69.
• › GPT-6.1 Sol: 10 million tokens a month ≈ $4.10; 100 million ≈ $40.97.
• › Gemini 2.5 Flash: 10 million tokens a month ≈ $6.49; 100 million ≈ $64.93.

Say a team routes 100 million tokens a month through Gemini 2.5 Flash on Quanta. As an example, that lands near $64.93 on the figures above. The same volume through GPT-6.1 Sol lands near $40.97. Either way, the number comes from published token prices rather than a negotiated quote, and every call in the dashboard shows the tokens and charge behind the total.

04 · What are the published limits on heavy workloads?

Concurrency and rate limits are part of Quanta's published and enforced limits, not a surprise discovered during a traffic spike. The limits are short enough to quote in full:

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

Streams carry no total-time cap. A stream ends only after 300 seconds of upstream silence, which suits long-running generation jobs where the answer arrives in pieces over an extended window.

For a team running several services at once, the combination matters. Twenty live keys allow work to be separated by service or environment, the 120 requests per minute per key gives each key its own throughput ceiling, and the 8 requests in flight per account keeps concurrency visible and controlled. Quanta publishes these numbers and enforces them, so capacity planning starts from documentation instead of trial and error.

05 · How do teams set up and settle the account?

Setup follows the protocol. Point an OpenAI-compatible client at the Quanta base address and use an OpenAI-compatible path for Kimi K3, GPT, DeepSeek, Qwen, Hunyuan, Grok and GLM models. Use the Anthropic native protocol with Claude Code or Claude Desktop for the Claude family, and the Gemini native protocol through Gemini CLI for Gemini models. One key works across all three.

Payment for companies runs by bank transfer from a company account. Invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard, so the paperwork follows the same place as the usage records. Quanta does not publish a support email address; the dashboard is where the conversation with the team happens.

Data handling is equally direct. The text users send is kept for 30 days for abuse handling and billing disputes, then deleted. Model answers are not stored. That distinction matters for teams that need to know what persists on the relay and what does not.

Put the pieces together and the case for Quanta is a practical one. One key and one balance serve three wire protocols. Token-level metering with per-call visibility replaces guesswork. Published limits of 120 requests per minute per key and 8 requests in flight per account make concurrency a known quantity. Prepaid pay-as-you-go billing, a $0.50 trial credit and refunds within 15 days keep the commitment proportional to use. For a team that runs heavy, sustained workloads and wants its rates, invoices and request history in plain sight, opening a Quanta account is a short decision to make.