Relay365Multi-model API relay desk
← All desk bulletins
Spend watchEN1 desk reads

Claude Code API Relay Pricing on Relay365: Cache Read at $0.2917

A handover note from the Relay365 duty desk

Welcome to the Relay365 duty desk. Relay365 is an AI API relay run like a round-the-clock duty desk, and it is an independent third-party relay, so everything here is about how we run the lanes and how we bill, not about anyone upstream. You are about to try the service, so this note is the one thing I want you to read before your first call lands.

The single habit that makes Relay365 easy to read is this: every lane sits on one table. Whatever model you wire up, the rate card, the usage log and the ledger all live in the same place, so you never have to open a second spreadsheet to understand what a call cost.

The second habit is that every request can be traced. The dashboard holds a usage log of the latest 200 requests, and each row shows the time, model, input, output, cache recall, charge and result of the call. That is where you will confirm the number this note is about.

Where cache recall sits on the Claude Fable 5 bill

On a Relay365 bill, four token types are priced separately: input, output, cache write and cache read. They are not blended into one line, which is the whole point of reading a bill instead of guessing. For Claude Fable 5, the wire name is claude-fable-5, and the rates per 1 million tokens in approximate USD are input $3.79, output $18.94, cache write $4.66 and cache read $0.2917.

Put those two numbers side by side and the picture is immediate: on Relay365, one million tokens of cache read on Claude Fable 5 costs $0.2917, while one million tokens of plain input costs $3.79. Cache recall is the cheap line because it reuses work you already paid to write, and the bill shows that as its own figure rather than hiding it inside the input rate.

Against official list prices, the facts we publish say Claude Fable 5 is listed at input $10 and output $50, so the Relay365 output rate is about 38% of official. That comparison is only for Claude and GPT lanes, where official list prices are quoted. Relay365 is an independent third-party relay, so treat those upstream numbers as reference points and treat the Relay365 rates as the ones your wallet settles on.

An example day of Claude Code calls on Relay365

Let me work one plain example so you know how to read your own log. Say you send 200,000 tokens of input and 2,000,000 tokens of cache read through Claude Fable 5 in a day. Input at $3.79 per 1 million tokens is 200,000 divided by 1,000,000 times $3.79, which is $0.758. Cache read at $0.2917 per 1 million tokens is 2,000,000 divided by 1,000,000 times $0.2917, which is $0.5834. Add them and the day costs $1.3414 for those two token types.

Notice what the example shows. The cache read side is several times larger in tokens than the input side, yet it costs less in dollars, because $0.2917 per million is a much lower rate than $3.79 per million. When you see that on a Relay365 row, you are looking at cache recall doing exactly the job you wired it for.

Every call is settled on the token counts the upstream reports, so the figures in the usage log are the figures that move the wallet. That is what makes a handover like this one short: you do not have to take my word for the arithmetic, you can read it back on the next screen.

How the wallet and the usage log keep the story straight

Relay365 runs on a prepaid wallet with no monthly fee, so there is no standing charge to reason about before you start. A call that fails upstream with a 502 or 503 is not billed. A disconnected stream is billed for what was actually used, which is the honest way to settle a call that got partway through.

The dashboard gives you API keys with tags and a key bound to a lane, quick-wiring snippets, the usage log of the latest 200 requests, and a ledger of wallet, bounty and pass accounts. There are in-site messages to the desk when you want to reach a human, and six interface languages. Note that the usage log does not show which key made the call, so plan your tags accordingly.

For a new account, there is a signup credit of $0.25, which is enough to make a few real calls and see a real bill before you commit anything else. If you top up and then change your mind, within 14 days of a top-up unused balance can be refunded minus what was consumed and the payment channel's fees, while bounty and promotional credit are not refundable.

When a duty pass makes sense instead of the wallet

Relay365 also offers duty passes, which are optional and cover Claude and GPT only; other models always use the wallet. A pass comes with a dedicated key that never touches the wallet, which keeps a long project's spend in one predictable place.

Take the Claude 14 day pass at $50.97 for 14 days. It carries $700 of official-price quota drawn at official price x2.5, which is about $280 of official-priced usage, with at most $380 of quota per 7 days. If your Claude Code work runs in bursts, that weekly ceiling is the number to plan around.

For a longer runway there is the Claude 30 day pass at $216.25 for 30 days, carrying $3200 of official-price quota drawn at official price x2.5, which is about $1280 of official-priced usage, with at most $900 of quota per 7 days. Both passes draw at the same x2.5 rate, so the difference is really about how much room you want and how long you want the key to stay live.

Wire it up the way the lane expects

Client rules on Relay365 are firm, so read them once and you will save yourself a confusing first hour. Claude models work only from the official Claude Code and Claude Desktop clients, Claude Code 2.1.285 or newer. They do not work from Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL.

Those other clients pair only with GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. Gemini runs on its native protocol only, through Gemini CLI, using :generateContent for single replies and :streamGenerateContent?alt=sse for streams, so it cannot go through /v1/chat/completions and it is not an OpenAI-compatible lane.

Two limits are worth knowing before a load test: an HTTP 429 comes back if you exceed more than 120 calls in 60 seconds on one key, or more than 8 requests in flight for one account. Stay inside those and the duty desk stays quiet. Create a Relay365 account and read your own cache recall on the first bill.

Further desk bulletins