RelayVano API Relay Pricing: Claude Code Top-Ups, Refunds and Balance

Top-ups on RelayVano: prepaid balance without a subscription

Most top-up mistakes start with a simple confusion: is a relay service selling subscriptions, or is it selling usage? RelayVano sells usage. You add funds to a prepaid balance, and every request draws down that balance by the tokens it actually consumes. There is no monthly fee, and the balance does not expire, so a quiet month does not force you to pay for a month you did not use.

Top-ups run through Stripe checkout. You can pay by card wherever cards are supported, and WeChat Pay is available for accounts where it is offered. When you first create an account, RelayVano starts you with $0.40 of credit, which is enough to send a test request and confirm that your key, your client and your chosen model are all talking to each other correctly.

A common mistake is topping up a large amount before checking how your workflow behaves. A smaller first top-up is easier to reason about, especially if you are still deciding between models. Another frequent error is assuming a prepaid balance behaves like a subscription you can cancel at the end of the month. It does not: you are paying per token as you go, and the unused portion stays on your balance.

RelayVano was built for developers worldwide, and the whole site, dashboard and docs come in six languages: 中文, English, Español, हिन्दी, Português, Français. Prices display in yuan on the Chinese pages and in approximate dollars on the others, while the charge itself is made in yuan. That makes it easier for a team spread across several countries to read the same documentation and compare the same numbers.

How the balance is charged, token by token

The second big mistake is treating the balance as a black box. RelayVano shows you what happened. Prompt tokens, reply tokens, cache stores and cache hits are priced separately, and every request appears with its charge on the usage page. Instead of guessing why the balance moved, you can open the usage record and see the cost attached to the request that caused it.

There is one billing detail worth internalising early. If a caller disconnects mid-stream, the request is still finished and billed on the tokens really used. In other words, closing a connection does not erase the work the model already performed. If your application retries aggressively after a timeout, those retries are separate requests and separate charges. A sensible retry policy is therefore cheaper than an impatient one.

Because RelayVano charges per token rather than per seat, the balance follows activity. A developer who ships one feature on a quiet week and another developer running many parallel requests will see very different balances at the end of the month, even on identical models. The usage page is the place to reconcile that difference.

One API key reaches every model we offer through https://api.relayvano.com, using the OpenAI-compatible /v1 path, the Anthropic /v1/messages path and the native Gemini /v1beta path. That single-key approach is what keeps setup short: you store the key once, point your tool at the base address, and choose the model you need for the task.

Client pairing: the mistake that breaks a setup fastest

The most expensive setup mistake has nothing to do with money. It is pointing the wrong client at the wrong model family. RelayVano supports specific pairings, and they are not interchangeable.

• Claude models work only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. If you are testing Claude from Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs or cURL, the request will not work as intended, because those clients pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models.
• Gemini models use the native Gemini protocol only, which means Gemini CLI. Gemini is not OpenAI-compatible, so treating it as an /v1 model is a protocol mistake rather than a configuration tweak.
• Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models.

Before you top up for a large test, send one small request through the client you actually plan to use. For example, if Claude Code is your daily tool, keep Claude there and verify the version is 2.1.285 or newer. If your team writes code through an OpenAI SDK, use that path with a GPT or DeepSeek model. Checking the pairing first costs one request; discovering it after a week of failed calls costs far more time.

Reading RelayVano pricing without fooling yourself

Fares below are approximate US dollars per million tokens, with 1 USD ≈ ¥7.2, and the charge itself is made in yuan. The number that matters is not the headline rate alone but the mix of prompt tokens, reply tokens and cache usage in your workload.

Consider an example. Suppose one request contains 1,000 prompt tokens and produces 500 reply tokens, with no cache activity. On Claude Sonnet 5.5, the prompt side is $1.25 per million tokens and the reply side is $6.25, so that single request costs roughly $0.004375. The same shape of request on MiniMax M3, at $0.1472 prompt and $0.5889 reply, costs roughly $0.00044165.

Now extend the example to 200 requests a day, which is an illustrative workload, not a measured figure. At that rate, Claude Sonnet 5.5 would come to about $0.875 per day, while MiniMax M3 would come to about $0.08833 per day. The point is not that one model fits every project. It is that a small, concrete calculation tells you more than a vague impression of a price list, and it tells you which model deserves a deeper look for your own traffic shape.

For comparison, RelayVano lists Claude Sonnet 5.5 at $1.25 prompt, $6.25 reply and $0.0889 cache hit, against an official list price of $2, $10 and $0.1 cache hit. GPT-6.1 Sol is listed at $0.1958 prompt, $0.9792 reply and $0.0389 cache hit, against an official list price of $2, $10 and $0.1 cache hit. GPT-5.6 Sol carries an OpenAI promotion stated to run at least until 2026-11-21, with RelayVano fares of $0.3736 prompt, $2.09 reply and $0.0306 cache hit against an official list price of $4, $20 and $0.4 cache hit. Compare the categories you actually generate, and remember that cache stores are priced separately from cache hits.

Refunds, referrals and getting help inside the dashboard

Refunds are straightforward within a defined window: unspent top-ups can be refunded within 9 days. The mistake to avoid is assuming a refund request can be made at any point, or that used funds are refundable. If you are still evaluating RelayVano, keep your first top-up modest so that part of it remains unspent inside the 9-day window.

Referrals work the same transparent way. You earn 16% of every top-up by people who sign up through your link, and 4% one level further. Earnings can be withdrawn from $5.56 or moved into your balance, so a small referral return can simply fund your next requests instead of being cashed out.

If anything about a top-up, a charge or a pairing is unclear, logged-in members write to us under Messages in the dashboard. That keeps support in the same place as the usage records, so you can point to the exact request and its charge rather than describing it from memory.

RelayVano combines a prepaid balance that does not expire, per-token billing you can inspect line by line, and a single key that reaches the models you need. Create your account, claim the $0.40 starter credit, send one test request through the client you use every day, and see how the usage page reports the charge. Then sign up and put RelayVano to work on your own traffic.