RelayVano Six-Language API Proxy: Claude Code, Gemini CLI, GPT

Pay-as-you-go instead of a subscription you rarely use

If you code with AI models only now and then, a vendor subscription can feel like the wrong shape of purchase. You pay a recurring amount for access, then spend much of the month not using it. RelayVano takes a different approach: you add money to a prepaid balance, and each request is charged for the tokens it actually uses. There is no monthly fee, and the balance never expires.

That arrangement is useful for a weekend script, a short debugging session, a one-off migration, or a month when your project needs light help rather than constant assistance. You are not committing to a fixed recurring payment for occasional coding. Instead, you decide how much to top up and draw from it as you work.

RelayVano is an independent relay made for developers worldwide. The entire site, dashboard, and documentation are available in six languages: 中文, English, Español, हिन्दी, Português, Français. Prices appear in yuan on the Chinese pages and in approximate dollars on the others, so you can read the information in a familiar format wherever you are.

New accounts start with $0.40 of credit. That is small, but it is enough to create an account, get familiar with the dashboard, and make your first request before deciding on a larger top-up. Support is also straightforward: logged-in members can write to us under Messages in the dashboard.

One key, several protocol families

RelayVano is designed around a simple idea: one API key reaches every model through our base address, api.relayvano.com. The key is the same, but the protocol you use depends on the model family you want to call.

For OpenAI-compatible models, the relay exposes an OpenAI-compatible /v1 endpoint. Anthropic-style requests use /v1/messages. Gemini models use the native Gemini /v1beta protocol, which is not OpenAI-compatible. That last point matters because your client must speak the protocol the model expects.

The pairing is deliberate and consistent:

• Claude models work only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer.
• Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs, and cURL pair only with GPT, DeepSeek, and the other non-Claude, non-Gemini models.
• Gemini models use the native Gemini protocol only, with Gemini CLI.

So if your daily tool is Claude Code, you can point it at RelayVano and use Claude models. If you prefer Cursor or the OpenAI SDKs, you can connect through the OpenAI-compatible route for GPT, DeepSeek, Qwen, MiniMax, Kimi, Grok, Hunyuan, and MiMo models. If you want Gemini, you use Gemini CLI and the native Gemini protocol. There is no need to treat the relay as a box of unrelated shortcuts; you choose the client and model combination that matches how you already work.

What the balance actually pays for

The billing model is based on tokens, and RelayVano separates the main cost categories: prompt tokens, reply tokens, cache stores, and cache hits. Each of those has its own rate. Every request also appears with its charge on the usage page, so you can see what happened rather than guessing from a single monthly total.

Consider a concrete example. Suppose you send 200 requests a day, and each request uses 1,000 prompt tokens and 500 reply tokens, with no cache activity. That is an example for illustration, not a typical usage profile. With MiniMax M3 at approximately $0.1472 per million prompt tokens and $0.5889 per million reply tokens, the daily prompt cost would be about $0.02944 and the daily reply cost about $0.29445, for roughly $0.32389 across 200 requests. Over 30 days, the same example would come to about $9.72.

Now try the same example with Qwen3.8 Max, at approximately $0.7431 per million prompt tokens and $2.23 per million reply tokens. The daily total would be about $1.60, or roughly $48 for 30 days. The example shows why model choice matters for anyone paying per request: a heavier model is still accessible when you need it, but it is worth checking the usage page before sending large batches.

Cache hits can change the arithmetic further, because they are priced separately and often at a lower rate. For instance, MiMo V2.6 Pro is listed at approximately $0.2319 for prompt tokens, $0.4653 for reply tokens, and $0.002778 for a cache hit. If your workflow can reuse cached context, that lower cache-hit rate may make sustained sessions more economical.

One detail worth knowing: if a caller disconnects mid-stream, the request is still finished and billed on the tokens really used. That behavior keeps accounting aligned with the work the model performed, even when the connection on your side was interrupted.

How RelayVano prices compare with official list prices

RelayVano is not affiliated with Anthropic, OpenAI, or Google. We are an independent relay, and our fares can be compared with the published list prices of the underlying vendors.

Take Claude Sonnet 5.5 as an example. RelayVano lists it at approximately $1.25 per million prompt tokens, $6.25 per million reply tokens, and $0.0889 for a cache hit. The official list price is $2, $10, and $0.1 for a cache hit. For a session with a large prompt and a long reply, that difference can add up over many requests.

GPT-6.1 Sol is listed at approximately $0.1958 for prompt tokens, $0.9792 for reply tokens, and $0.0389 for a cache hit, compared with an official list price of $2, $10, and $0.1. GPT-5.6 Sol is listed at approximately $0.3736, $2.09, and $0.0306, against an official list price of $4, $20, and $0.4. The official promotion for GPT-5.6 Sol is stated to run at least until 2026-11-21.

These figures are approximate US dollars per million tokens, using an exchange rate of about 1 USD ≈ ¥7.2, while the actual charge is made in yuan. RelayVano does not sell subscriptions. We sell pay-as-you-go balance, and the rates above show what drawing that balance costs.

Paying, refunding, and referring

Top-ups run through Stripe checkout. Cards work everywhere, and WeChat Pay is available where your account supports it. If you change your mind about an unused top-up, unspent top-ups can be refunded within 9 days. That gives you room to try RelayVano without turning a small experiment into a long-term commitment.

You can also share the service and earn from the top-ups it brings. People who sign up through your link generate 16% of every top-up for you, with 4% coming one level further. Earnings can be withdrawn from $5.56 or moved into your RelayVano balance, where they are available for requests straight away.

The practical flow is short. Create an account, use the $0.40 starter credit if you want, choose a model family, and point the matching client at api.relayvano.com with your key. When you are ready for more, add funds through checkout and watch the usage page record each charge. Questions along the way go to Messages in the dashboard.

When pay-as-you-go fits your coding routine

Pay-as-you-go is a good fit when your usage is uneven. It also fits when you want access to more than one model family without managing several separate subscriptions. With RelayVano, one key reaches GPT, DeepSeek, Claude, Gemini, Qwen, Kimi, MiniMax, Grok, Hunyuan, and MiMo models, and you choose the client that matches the model rather than forcing a model into the wrong tool.

The six-language site, dashboard, and documentation were built for developers worldwide, so the setup does not assume you read English or browse from one country. Pricing shown in yuan or approximate dollars, Stripe checkout with cards, and WeChat Pay where supported give you familiar ways to fund the balance.

What remains is a simple decision. If a subscription would sit mostly unused, compare its recurring cost with the token-based fares on your actual requests. Start with the $0.40 credit, send a few calls, read the charges on the usage page, and then top up the amount you are comfortable with. The balance does not expire, and there is no monthly fee to remember.

Ready to try it? Sign up for RelayVano, add your key, and take your first request from there.