Claude Code API relay pricing at Relay365 for freelancers
The first key of the morning
Picture a small project landing in your inbox on a Tuesday: a client wants a summarisation tool, a batch of classification calls, and a weekly report, all riding on one model family. The first thing you need is a key that works at 2 a.m. exactly as it does at noon. Relay365 is run like a round-the-clock duty desk: an independent third-party relay that keeps every lane on one table, behind a single API base, with in-site messages to the desk whenever you log in. A new account starts with a $0.25 signup credit, so the first test call costs almost nothing and the ledger shows you precisely what it cost.
Because Relay365 settles every call on the token counts the upstream reports, the freelancer's hardest question, what did this actually cost, has a plain answer before any invoice goes out. There is no monthly fee, no subscription to cancel, and the wallet is prepaid: you top up, you spend, and the balance is yours.
Every lane on one table
One project rarely means one model. On Relay365 you might draft with Claude Haiku 4.5 at $0.3542 input and $1.70 output per 1 million tokens, run the heavy lifting on Claude Sonnet 5.5 at $1.18 input and $5.91 output, push high-volume jobs through DeepSeek V4 Flash at $0.2889 input and $1.15 output, and serve OpenAI-compatible clients through GPT-6 Astra at $0.7444 input and $3.72 output. All rates are per 1 million tokens in approximate USD while the account itself is kept in CNY, and the full table sits on the pricing page.
The desk keeps firm client rules so nothing breaks mid-project. Claude models answer only from the official Claude Code and Claude Desktop clients, Claude Code 2.1.285 or newer; they do not run from Cursor, Cherry Studio, Cline, OpenCode, any SDK or cURL, and those tools pair only with the GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes. Gemini runs on its native protocol only, through Gemini CLI, with :generateContent for single replies and :streamGenerateContent?alt=sse for streams, never through /v1/chat/completions.
A usage line for every request
This is where Relay365 earns its keep for billing. The dashboard keeps a usage log of the latest 200 requests, and each line shows the time, the model, the input tokens, the output tokens, the cache recall, the charge and the result of that one call. When your client asks why a line item exists, you can point at the log rather than at a guess. A call that fails upstream with a 502 or 503 is not billed at all, and a disconnected stream is billed only for what was actually used.
Each line also separates the four token types, because they are priced separately: input, output, cache write and cache read each carry their own rate. On Claude Sonnet 5.5, for example, cache read costs $0.0944 per 1 million tokens against $1.18 for fresh input, so a project that reuses context spends visibly less, and the log shows that saving line by line.
Working the numbers on one small project
Sums help, so here is one, worked only from the published rates. For example, if you send 2 million input tokens and 500,000 output tokens in a day on claude-sonnet-5-5, the day costs 2 × $1.18 plus 0.5 × $5.91, which is $2.36 plus $2.955, or $5.315 for the day. If 1 million of those input tokens come back as cache reads instead of fresh input, that portion costs $0.0944 per million rather than $1.18, a saving you can quote to the client. For comparison, the Relay365 output rate on this model is about 59% of the official list price, and on GPT-6 Astra the output rate is about 7% of official.
Keys make this traceable per project too: the dashboard issues API keys with tags and a key bound to a lane, with quick-wiring snippets beside them, so one client's key never mixes with another's spend even though the usage log itself does not show which key made the call.
From the log to the first bill
Billing stays simple because the wallet is prepaid and every call is settled on the token counts the upstream reports. If a top-up turns out to be too large, unused balance can be refunded within 14 days of the top-up, minus what was consumed and the payment channel's fees; bounty and promotional credit are not refundable. If the project grows, an optional duty pass can carry the Claude or GPT traffic on a dedicated key that never touches the wallet: for example, the Claude 14-day pass costs $50.97 and draws up to $700 of quota at the duty-pass rate, with at most $380 of quota per 7 days. The GPT 30-day pass at $81.53 works the same way with up to $3000 of quota and at most $900 per 7 days.
Two housekeeping notes from the desk: a key trips an HTTP 429 only above 120 calls in 60 seconds, or more than 8 requests in flight for one account, so a small project will never brush the ceiling; and if you bring other freelancers aboard, the referral bounty pays 18% at level 1 and 6% at level 2 of what referred members pay, withdrawable from $11.11 or movable into the wallet.
Closing the shift
The shift log of this small project reads like a clean ledger: first key in the morning, every lane on one table, a usage line for every request, four token types priced apart, and a first bill that matches the log to the cent. Relay365 keeps the desk staffed around the clock so the next project starts the same way. Create a Relay365 account and wire your first key tonight.