Claude Code and an API relay: Relay365 on one prepaid table

The first telling of your working day: vendor tabs and scattered balances

Here is a working day, told twice. In the first telling, you are a solo developer sitting down before the first commit with six browser tabs open and three different provider dashboards behind them. One tab holds a balance you topped up last week. Another holds a key you have to copy and paste into a side project. A third reminds you that the model you wanted for a long-context refactor is not the model you already configured, and the invoice for the other one arrives separately. Every provider has its own wallet, its own key format, its own dashboard, and its own idea of what a request costs. You become the integration layer, spending your build hours on plumbing instead of product.

Relay365 is the second telling of the same day. It is an independent third-party relay run like a round-the-clock duty desk: every lane sits on one table, and one prepaid wallet covers the whole table. There is no monthly fee. Rates are quoted per 1 million tokens in approximate USD, while the account itself is kept in CNY, so a new account can start with $0.25 in signup credit and a clear view of what a call will cost before you commit a single token. The duty desk at Relay365 is there to keep the lanes running and the ledger readable while you get on with your own work.

In the second telling, the tabs are gone. You open one dashboard, create API keys, tag them, and bind each key to the lane it is meant to serve. Quick-wiring snippets sit next to the key list, so wiring a new lane does not become an afternoon of reading scattered documentation. You still choose models deliberately, because different lanes suit different jobs, but the choice happens inside one account instead of across several. The same work, the same tokens, and a much shorter path from idea to shipped feature.

What the lanes cost when the arithmetic is done in plain sight

Relay365 prices the four token types separately: input, output, cache write and cache read. That separation matters because a request is not one price but four moving parts, and a relay that only shows a blended number leaves you guessing. The rates below are the ones the desk works from, per 1 million tokens in approximate USD.

Take Claude Fable 5 on the wire as claude-fable-5: input $3.79, output $18.94, cache write $4.66 and cache read $0.2917. Its output rate is about 38% of the official list price of $50. For example, if you send 1 million input tokens and receive 1 million output tokens a day, that is $3.79 plus $18.94, or $22.73 a day and $681.90 over 30 days. Cache reads are cheaper still at $0.2917 against an official list price of $1, so a workflow that leans on cached context keeps its meter readable.

Now take GPT-6.1 Sol on the wire as gpt-6.1-sol: input $0.1986, output $0.9944 and cache read $0.0389. Its output rate is about 10% of the official list price of $10. For example, if you send 1 million input tokens and receive 1 million output tokens a day, that is $1.193 a day and $35.79 over 30 days. For lighter lanes, DeepSeek V4 Flash is input $0.2889, output $1.15 and cache read $0.006944, so the same 1 million in and 1 million out a day comes to $1.4389, or $43.17 over 30 days. Gemini 2.5 Flash is input $0.2389, output $1.91 and cache read $0.0236, which makes the same daily pattern $2.1489, or $64.47 over 30 days. No official price is quoted for the DeepSeek or Gemini families, so the desk makes no comparison for them; the rates stand on their own. The full table on the pricing page shows every lane.

The desk keeps the log, so every request can be traced

Because the four token types are priced separately, they are also recorded separately. The Relay365 usage log shows the latest 200 requests with the time, the model, input, output, cache recall, the charge and the result of each call. One table, one place to look when a bill feels heavier than expected. The log does not show which key made a call, so if you need that level of detail, the tagged keys and the lane binding are where you start.

Beside it sits a ledger that separates the wallet, bounty and pass accounts, so promotional money, referral earnings and your own top-ups never blur into one confusing number. In-site messages go straight to the desk, though you need to log in first. The interface is available in six languages. Together, these are small details that decide whether a relay feels like a tool or like another chore: the duty desk at Relay365 is meant to be checked, not decoded.

Wiring the lanes to the right clients

Client rules at Relay365 are firm, and the desk would rather explain them once than debug them for you. Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Cursor, Cherry Studio, Cline, OpenCode, any SDK and cURL pair only with the GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. So the split is simple: heavy Claude work happens in the Claude clients, while everything else can be wired into the tools you already use.

Gemini runs on its native protocol only, through Gemini CLI. A single reply goes through :generateContent, and a stream goes through :streamGenerateContent?alt=sse. It cannot go through /v1/chat/completions. That is not a limitation to work around; it is the shape of the lane, and Relay365 hands it to you cleanly rather than pretending it is OpenAI-compatible.

There are guardrails worth knowing. An HTTP 429 appears when a single key makes more than 120 calls in 60 seconds, or when one account has more than 8 requests in flight. And billing follows what actually happened: every call is settled on the token counts the upstream reports, a call that fails upstream with a 502 or 503 is not billed, and a disconnected stream is billed only for what was actually used. No mystery charges from a call that never completed.

Duty passes, wallet rules and a clean exit

If a build sprint needs more room than a normal day, Relay365 offers duty passes for Claude and GPT only. The Claude 14-day pass costs $50.97 for 14 days and draws up to $700 of quota at the duty-pass rate x2.5, with at most $380 of quota per 7 days; at list prices that is at most about $280 of usage, and less on models whose pass rate sits above list. The GPT 30-day pass costs $81.53 for 30 days and draws up to $3000 of quota at the duty-pass rate x2.5, with at most $900 of quota per 7 days, which is at most about $1200 of usage at list prices and less where the pass rate sits above list. A pass comes with a dedicated key that never touches the wallet, and other models always use the wallet.

Your money stays legible too. A top-up can be refunded within 14 days, with the unused balance returned minus what was consumed and the payment channel's fees; bounty and promotional credit are not refundable. And if you bring a colleague along, the referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, withdrawable from $11.11 or moved into the wallet.

That is the whole shift, really. One table instead of six tabs, one balance instead of three, one log that shows what every request cost, and a desk that answers when something needs explaining. If you are ready to work that way, create a Relay365 account today and start with $0.25 in signup credit on the duty desk.