Claude Code API relay pricing: where Relay365 logs each request
Where the duty desk shows what a request cost
People ask the Relay365 desk a simple question: where can I see what each request actually cost? The answer is the usage log in the Relay365 dashboard, and it is deliberately plain. It keeps the latest 200 requests, and for each one it shows the time, the model, the input, the output, the cache recall, the charge and the result of the call. That is the desk's answer to a billing question rather than a marketing page: you can read back what happened on any single request instead of estimating from a monthly total.
There is one honest limitation worth stating at the start. The usage log does not show which key made the call. If your team runs several keys from one account, the log tells you the model, the tokens and the charge, but you will need your own tags and your own notes to attribute a request to a project. Relay365 keeps API keys with tags and lets you bind a key to a lane, which helps you organise traffic before it happens; it does not rewrite the log to add a key column that is not there.
Alongside the log, the dashboard carries a ledger of wallet, bounty and pass accounts, so the money you have available and the money you have spent sit in the same place. There are also quick-wiring snippets for pointing a client at the Relay365 base URL, and in-site messages to the desk once you are logged in. The whole point of the desk is that you should not have to send a message to find out where a number came from.
Four token types, priced on four separate lines
Relay365 prices four token types separately, and the usage log reports them separately too: input, output, cache write and cache read. That separation is the most useful thing to understand about a Relay365 bill, because a request that looks inexpensive on input can still cost on output, and a request that writes to the cache is not the same as one that reads from it.
Take Claude Opus 5.5, whose wire name is claude-opus-5-5. Per 1 million tokens in approximate USD, input is $1.61, output is $7.59, cache write is $1.90 and cache read is $0.0736. The official list price for the same model is input $4, output $20 and cache read $0.2, so the Relay365 output rate is about 38% of official. Read that as a comparison of rates, not as a promise about a bill: your actual spend depends on the token mix your application produces.
GPT-6.1 Sol, on the wire gpt-6.1-sol, is listed at input $0.1986, output $0.9944, no cache write line, and cache read $0.0389 per 1 million tokens. The official list price is input $2, output $10 and cache read $0.1, so the Relay365 output rate is about 10% of official. Gemini 3.1 Pro, the wire name gemini-3.1-pro, is listed at input $1.54, output $8.94, cache write $1.54 and cache read $0.1542. No official price is quoted for that family, so the desk does not invent one to compare against.
The trade-off is straightforward. Caching is where the accounting gets interesting: cache read is a fraction of input on every model quoted here, so a workload that reuses context benefits, while a workload that writes fresh context every turn pays the cache write line. If you are unsure which pattern your code has, the usage log is where the answer surfaces, because cache recall is reported per request.
What a day of traffic looks like on the Relay365 wallet
Relay365 runs on a prepaid wallet. There is no monthly fee, and every call is settled on the token counts the upstream reports. 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 a fair but occasionally unwelcome detail: a stream that dies halfway still shows a charge for the part that arrived.
For example, if you send 1 million input tokens and 100,000 output tokens of Claude Opus 5.5 a day, that is $1.61 plus $0.759, or $2.369 a day. The same daily shape on GPT-6.1 Sol is $0.1986 plus $0.09944, or $0.29804 a day. On Gemini 3.1 Pro it is $1.54 plus $0.894, or $2.434 a day. These are examples worked out from the published rates, not quotations, and they leave out cache traffic entirely.
Money moves in and out with a few rules worth reading once. A new account receives $0.25 in signup credit. The referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, and it can be withdrawn from $11.11 or moved into the wallet. Refunds are available 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.
The account is kept in CNY, while the rates are quoted in approximate USD per 1 million tokens. That is a small piece of friction worth naming, and it is part of why the desk keeps the per-request log in view: when the account currency and the rate sheet currency differ, the usage log is where the per-request charge in the rate sheet's terms is visible.
Duty passes and the trade-offs they bring
Claude and GPT also have an optional duty pass, and this is where the honest trade-offs are sharpest. A pass comes with a dedicated key that never touches the wallet, and 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, with the overrun counting against the earliest-expiring card. 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.
Consider the Claude pass for 30 days at $720, which costs $94.03 for 30 days and allows up to $720 of quota drawn at the duty-pass rate x2.5. That is at most about $288 of usage at list prices, and less on models whose pass rate sits above list, with at most $230 of quota per 7 days. The GPT pass for 30 days at $2350 costs $81.53 for 30 days, allows up to $2350 of quota drawn at the duty-pass rate x2.5, at most about $940 of usage at list prices, and caps quota at $700 per 7 days.
So what do you give up? The pass rate is x2.5 the duty-pass rate, so a pass is not a discount on the wallet rates; it is a prepaid block of capacity at a different multiplier, with its own weekly ceiling and its own concurrency behaviour. The gain is that the pass key never touches the wallet, so a busy week draws on a fixed quota instead of your balance. Other models always use the wallet, and a pass is offered for Claude and GPT only.
The desk's suggestion is to start on the wallet if your volume is uncertain, and to consider a pass when you already know the shape of your traffic and want the weekly ceiling to do some of the planning. The pricing page shows the full table of passes and rates, so you can check the arithmetic against your own numbers before you commit.
Client rules and limits that shape your setup
Client rules are firm at Relay365, and they affect how you will see your costs. 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, SDK clients and cURL pair only with the GPT, DeepSeek and other non-Claude lanes. Gemini runs on its native protocol only, through Gemini CLI: :generateContent for a single reply and :streamGenerateContent?alt=sse for a stream, and it cannot go through /v1/chat/completions.
Rate limits are plain as well. You will see HTTP 429 if one key makes more than 120 calls in 60 seconds, or if one account has more than 8 requests in flight. Those limits shape how your usage log looks from day to day: a burst that trips the limit shows up as a rejected call rather than a slow one.
Pulling the threads together, the trade-offs are real. The usage log shows the latest 200 requests and does not attribute them to a key. Passes multiply the rate by x2.5 and impose weekly ceilings. Streams that disconnect are billed for what was used. What you get in return is one table for every lane, per-request traceability across input, output, cache write and cache read, a prepaid wallet with no monthly fee, and published rates that put a Claude Opus 5.5 output call at about 38% of official and a GPT-6.1 Sol output call at about 10% of official.
That is the deal the duty desk is offering: transparent per-request numbers, clear limits, and rates that leave room in the budget. When you are ready, create a Relay365 account and put your own traffic on the log.