Claude Code pricing: a student API relay bill with Relay365

Why students want the bill in front of them

The duty desk at Relay365 runs like a round-the-clock duty desk, and our job is simple: keep every lane on one table, price every token type separately, and leave a trail that a student can actually read. Relay365 is an independent third-party relay with a prepaid wallet and no monthly fee. Every call is settled on the token counts the upstream reports, so the number you see on the bill is the number the model service sent back.

Four token types are priced on their own: input, output, cache write and cache read. That separation is what makes a small bill legible. Instead of one blended figure, you can see which part of your prompt cost money, which part cost more, and what caching did for you. The dashboard keeps a usage log of the latest 200 requests with the time, model, input, output, cache recall, charge and result of each call, so nothing has to be taken on faith.

The worked figures below are examples. They are not averages, forecasts or promises about your own project; they are simply sample volumes worked out from the published rates, so you can see the method before you top up.

A worked Claude Fable 5.1 bill, line by line

Claude Fable 5.1, on the wire name claude-fable-5-1, is priced per 1 million tokens in approximate USD: input $4.00, output $20.01, cache write $4.93 and cache read $0.2375. The account itself is kept in CNY, so the display and the settlement follow your account settings while the rate card below stays in these approximate USD figures.

For example, if you send 100,000 input tokens, 20,000 output tokens, 50,000 cache write tokens and 200,000 cache read tokens in a day, the bill for that day is:

• Input: 100,000 tokens is 0.1 million, so 0.1 × $4.00 = $0.40.
• Output: 20,000 tokens is 0.02 million, so 0.02 × $20.01 = $0.4002.
• Cache write: 50,000 tokens is 0.05 million, so 0.05 × $4.93 = $0.2465.
• Cache read: 200,000 tokens is 0.2 million, so 0.2 × $0.2375 = $0.0475.
• Day total: $0.40 + $0.4002 + $0.2465 + $0.0475 = $1.0942.
• Thirty days at the same volume: $1.0942 × 30 = $32.826, which is about $32.83.

Now try a heavier day in the same way. For example, if you send 300,000 input tokens, 60,000 output tokens, 100,000 cache write tokens and 500,000 cache read tokens in a day:

• Input: 0.3 × $4.00 = $1.20.
• Output: 0.06 × $20.01 = $1.2006.
• Cache write: 0.1 × $4.93 = $0.493.
• Cache read: 0.5 × $0.2375 = $0.11875.
• Day total: $1.20 + $1.2006 + $0.493 + $0.11875 = $3.01235.
• Thirty days at that pace: $3.01235 × 30 = $90.3705.

The published official list price for this model is input $10, output $50 and cache read $0.25, so the Relay365 output rate is about 40% of official. On the heavier example above, the same input, output and cache read volumes at official list prices would be $3.00 + $3.00 + $0.125 = $6.125; the cache write line is excluded because no official cache write price is quoted for this family. The comparison is offered only where a published price exists, and the pricing page shows the full table.

The same arithmetic on three other lanes

The method does not change when you switch lanes, which is the point of putting every lane on one table. Claude Fable 5, on the wire name claude-fable-5, charges input $3.79 and output $18.94 per 1 million tokens. For example, if you send 1,000,000 input tokens and 200,000 output tokens, that is $3.79 + $3.788 = $7.578. The official list price for this model is input $10 and output $50, and the Relay365 output rate is about 38% of official, so the same two lines at official list prices would be $10 + $10 = $20.

Claude Haiku 4.5, on the wire name claude-haiku-4-5-20251001, is priced at input $0.3542 and output $1.70 per 1 million tokens, with the Relay365 output rate about 34% of official. For example, if you send 500,000 input tokens and 100,000 output tokens, that is 0.5 × $0.3542 = $0.1771 plus 0.1 × $1.70 = $0.17, for $0.3471 in total. At the official list price of input $1 and output $5, the same volumes would be $0.50 + $0.50 = $1.00.

Gemini 2.5 Flash, on the wire name gemini-2.5-flash, is priced at input $0.2389, output $1.91, cache write $0.2389 and cache read $0.0236 per 1 million tokens. For example, if you send 1,000,000 input tokens and 500,000 output tokens, that is $0.2389 + $0.955 = $1.1939. No official price is quoted for this family, so no comparison is drawn here. Gemini runs on its native protocol only, through Gemini CLI, using :generateContent for single replies and :streamGenerateContent?alt=sse for streams, and it cannot go through /v1/chat/completions.

A small top-up or a duty pass

For the light example above, a top-up of roughly $33 would cover a month of that habit on the wallet, and a new account also receives $0.25 in signup credit. If you would rather not watch a wallet balance, Relay365 also offers duty passes for Claude and GPT. A pass comes with a dedicated key that never touches the wallet, and quota is drawn at the duty-pass rate x2.5.

Two Claude passes illustrate the choice. The 14-day pass costs $50.97 and carries up to $700 of quota drawn at the duty-pass rate x2.5, which is at most about $280 of usage at list prices and less on models whose pass rate sits above list, with at most $380 of quota per 7 days. The 30-day pass costs $94.03 and carries up to $1400 of quota, at most about $560 of usage at list prices and less on models whose pass rate sits above list, with at most $450 of quota per 7 days. On the worked monthly wallet example, $32.826 is a smaller outlay than either pass, so the pass is worth it when your usage is heavier and steadier than the sample above.

Refunds stay straightforward: 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. A call that fails upstream with a 502 or 503 is not billed, and a disconnected stream is billed for what was actually used.

Following the request on the dashboard

Every request can be traced in the usage log, which shows the time, model, input, output, cache recall, charge and result of each call for the latest 200 requests. It does not show which key made the call, so when several keys are in play, the ledger of wallet, bounty and pass accounts is where the balances live. API keys can carry tags, and a key can be bound to a lane, which keeps a classroom project from wandering into a model you did not mean to call. Quick-wiring snippets are on the dashboard, in-site messages reach the desk once you log in, and the interface is available in six languages.

There are limits worth knowing before a deadline week. HTTP 429 appears when there are more than 120 calls in 60 seconds on one key, or more than 8 requests in flight for one account. Client rules are firm as well: Claude models work only from the Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Cursor, Cherry Studio, Cline and OpenCode pair only with the GPT, DeepSeek, Kimi, GLM, Qwen, Grok, MiniMax and Hunyuan lanes.

Referrals work in your favour too. The bounty is 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. The desk keeps every lane on one table, prices the four token types separately, and writes down each charge so the bill can be checked line by line. When the worked example above matches the way you actually use models, create a Relay365 account and put your own first request on the log.