Relay365 relay service pricing for Claude Fable 5.1 tokens

Why the night desk keeps a ledger for you

Dear developer on the late shift, this is the Relay365 duty desk writing to you about the thing that quietly eats a weekend: money spent on AI calls you cannot account for. You open a dashboard, see a number, and wonder which project paid for it, which key sent it, and whether the cache made it cheaper or simply moved the bill. The Relay365 ledger exists to answer those questions without a spreadsheet archaeology session.

Relay365 is an AI API relay run like a round-the-clock duty desk, and it is an independent third-party relay. Every lane sits on one table, so the same page holds the wallet account, the bounty account and any duty pass you are holding. You are not juggling a separate console for each model family; you are standing in front of one desk, and the desk keeps the receipts.

The order of business is simple. Money goes into a prepaid wallet, calls draw from it, and every request can be traced in the usage log. There is no monthly fee. Each call is settled on the token counts the upstream reports, which means the figure on the log is the figure that was billed. If a call fails upstream with an HTTP 502 or 503, it is not billed at all. If a stream disconnects halfway, you are billed for what was actually used, not for the whole intended response.

That is the whole promise of the ledger, and it is why the desk stays staffed: clarity you can check line by line.

Wallet, bounty and the four separate token prices

The wallet is the account you fund and spend from. The bounty account is where referral rewards land: 18% at level 1 and 6% at level 2 of what referred members pay. You can withdraw a bounty from $11.11, or move it into your Relay365 wallet and let it work for you. A new account also starts with $0.25 in signup credit, a small cushion while you wire the first key.

Every lane is priced on four token types, and the ledger prices each one separately: input, output, cache write and cache read. This is the detail that ends most arguments about a bill. On Claude Fable 5.1, for example, input is $4.00 per 1 million tokens, output is $20.01, cache write is $4.93 and cache read is $0.2375. The output rate is about 40% of the official list price of $50.

Here is a worked example, and it is only an example: if you send 1,000,000 input tokens, 200,000 output tokens and 1,000,000 cache read tokens of Claude Fable 5.1 in a day, that is $4.00 for input, $4.002 for output and $0.2375 for the cache read, or about $8.24 for the day. Same table, different lane, different shape.

On GPT-6.1 Sol the same arithmetic lands much lower: input is $0.1986 per 1 million tokens, output is $0.9944 and cache read is $0.0389, and the output rate is about 10% of the official list price of $10. For example, if you send 1,000,000 input tokens and 1,000,000 output tokens of GPT-6.1 Sol a day, that is $0.1986 plus $0.9944, or about $1.19. Gemini 2.5 Flash is priced on the same four columns too, at $0.2389 input, $1.91 output, $0.2389 cache write and $0.0236 cache read per 1 million tokens.

Refunds are handled plainly. Within 14 days of a top-up, unused balance can be refunded minus what was consumed and the payment channel's fees. Bounty and promotional credit are not refundable, so treat them as spend rather than cash. Everything else is visible in the ledger, and the pricing page carries the full table for every lane.

Reading the usage log line by line

The usage log is the part developers check first. It holds the latest 200 requests, and each row shows the time, the model, input, output, cache recall, the charge and the result. One honest note: the log does not show which key made the call, so if you need per-key attribution, use tags and lane binding to keep your own naming discipline.

That is why the Relay365 dashboard also lets you create API keys with tags and bind a key to a lane. A key tagged for one lane and another tagged for another keeps your experiments apart before the log ever sees them. Quick-wiring snippets sit on the same page, so a new lane can be wired without a scavenger hunt through documentation.

The ledger sits alongside the log, listing your wallet, bounty and pass accounts in one place, and in-site messages let you reach the desk directly once you log in. The interface is offered in six languages. If a number looks wrong, the message thread is the shortest path to a person reading the same ledger you are.

Duty passes for Claude and GPT lanes

Some nights you would rather not meter tokens at all. Relay365 duty passes are the optional answer, and they are available for Claude and GPT only; other models always use the wallet. A pass comes with a dedicated key that never touches the wallet, so pass traffic and wallet traffic cannot muddy each other.

The Claude pass for 30 days carries $720 of quota with a $230 per week ceiling, and it costs $94.03 for 30 days. The GPT pass for 30 days carries $2350 of quota with a $700 per week ceiling, and it costs $81.53 for 30 days. Both draw at the duty-pass rate x2.5. On the Claude pass, that is at most about $288 of usage at list prices, less on models whose pass rate sits above list; on the GPT pass, at most about $940 of usage at list prices.

One rule is worth reading twice: a call already let through on a pass is drawn in full from the pass, even if it runs past the quota or the 7-day ceiling, and the overrun counts against the earliest-expiring card. The Relay365 desk does not stop a call mid-flight and does not quietly move the cost to your wallet.

When a pass is nearly used up, each running call holds $3 x the draw rate, and an extra parallel call gets HTTP 429 until the running ones finish. That is the desk protecting the quota you paid for, and it is also a reminder not to fan out more parallel work than your pass can carry.

Client rules and the limits the desk keeps

Before you wire anything, the client rules matter. Claude models work only from the official Claude Code and Claude Desktop clients, and Claude Code 2.1.285 or newer is required. Cursor, Cherry Studio, Cline, OpenCode, any SDK and cURL pair only with the GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes.

Gemini runs on its native protocol only, through Gemini CLI: :generateContent for single replies and :streamGenerateContent?alt=sse for streams. It cannot go through /v1/chat/completions, so a Gemini lane is a different wiring job rather than a copy of the others.

The desk also keeps a short list of limits, because a quiet throttle is easier to plan around than a surprise failure. An HTTP 429 arrives when one key sends more than 120 calls in 60 seconds, or when one account keeps more than 8 requests in flight.

And if you want to start tonight, create a Relay365 account, fund the wallet, and tag your first key to the lane you plan to use. The ledger will keep the rest straight.