Claude Fable 5.1 API relay limits for high-volume teams

01 · What Quanta publishes about limits on heavy traffic

When a team runs production traffic through an API relay, the most useful documentation is the one that states the ceilings in plain numbers. Quanta publishes and enforces its limits, so you can size a workload against them before pressure builds. Quanta allows 120 requests per minute per key, with 8 requests in flight per account, a request body ceiling of 25 MB and up to 20 live keys per account. These are not advisory figures; they are the boundaries Quanta applies.

Heavy, sustained workloads strain predictable places. A batch job that fires faster than the per-minute ceiling runs into a published boundary, and a service that opens more concurrent calls than the in-flight allowance runs into another. Knowing the numbers in advance lets a team decide where to add a key, where to pace a worker and where a long-running task belongs. Quanta gives your team the figures; the architecture stays yours.

02 · Streams, long responses and the 300-second silence rule

Heavy traffic is not always a burst of short calls. Quanta's streaming behaviour is stated just as explicitly: streams carry no total-time cap, and they end only after 300 seconds of upstream silence. A long generation is therefore not cut off simply because it ran long; it ends when the upstream stops sending for five minutes. For teams running sustained jobs, that removes one more unknown from the operating picture.

Across the three protocols Quanta supports — OpenAI-compatible at /v1, Anthropic native and Gemini native — you use one key and one balance. That is a practical answer to a common relay question: a workload that spans model families does not need a separate account, a separate credit pool or a separate set of credentials for each protocol. Quanta keeps the billing surface consolidated while the published ceilings keep traffic predictable.

One key can carry 120 requests per minute, and an account can hold 20 live keys. For teams that want separation between services or environments, the key count provides room to organise traffic without splitting the balance. It is a deliberately small, published set of controls, and it is enforced as written.

03 · What Claude Fable 5.1 costs on Quanta, token by token

Quanta meters every call on real tokens. Input, output, cache writes and cache reads are priced separately, and the dashboard lists the tokens and the charge for every call, so a bill can be traced back to individual requests rather than to an estimate.

Claude Fable 5.1 on Quanta is listed at $4.38 per 1M input tokens, $21.88 per 1M output tokens, $5.39 per 1M cache write tokens and $0.2222 per 1M cache read tokens. These are approximate USD figures; charges are made in CNY. Because cache reads are priced separately and shown on the dashboard, a team that reuses context can see that effect in the per-call figures instead of discovering it in a summary.

Within the same published list, Claude Opus 5.5 is $1.79 / $8.46 / $2.11 / $0.0819 and Claude Sonnet 5.5 is $1.25 / $6.28 / $1.57 / $0.0917. Quanta also lists GPT-6.1 Sol at $0.2083 / $1.01 / - / $0.0403, Gemini 2.5 Flash at $0.2361 / $1.89 / $0.2361 / $0.0236, DeepSeek V4 Pro at $1.44 / $4.32 / - / $0.0486, and Qwen3.7 Plus at $0.1667 / $0.4861 / - / $0.0486.

To make the scale concrete, consider these examples 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 runs about $25.07 and $250.69 on the same two volumes. These are examples, not quotes, but they show how Quanta's per-token metering behaves as a workload grows.

04 · Where Claude Fable 5.1 runs, and where other models run

Claude Fable 5.1 is reached through the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Claude models work only in those two clients. They do not pair with Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs or cURL, so if your team standardises on one of those tools, the Claude line is not the route to plan around.

Gemini models use the native protocol only, through Gemini CLI; Gemini is not served through the OpenAI-compatible path. Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs and cURL pair with GPT, DeepSeek and the other non-Claude, non-Gemini models on Quanta's list, including GPT-6 Astra, Qwen3.8 Max, Kimi K3, Hunyuan 4, Grok 4.7 and GLM-5.3.

For a team running several toolchains, Quanta's one-key, one-balance model means those routes share the same prepaid funds and the same dashboard, while the client rules stay unambiguous.

05 · Prepaid balance, top-ups, refunds and company invoices

Quanta is pay-as-you-go. There is no monthly fee and no minimum spend; you fund a prepaid balance and calls draw down from it as tokens are metered. Top-ups are available at $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89, so a team can start small and scale its balance in line with real usage. New accounts receive $0.50 in trial credit on sign-up.

Unused balance can be refunded within 15 days of a top-up. 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.

For company buyers, payment is by bank transfer from a company account, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. Quanta does not bury the commercial terms behind a sales queue: the limits, the token prices and the invoice route are all stated.

On data handling, the text users send is retained for 30 days for abuse handling and billing disputes, then deleted; model answers are not stored.

Heavy traffic needs a relay that publishes its ceilings and meters its charges. Quanta does both, and the next step is a funded balance and a key that fits your workload.