RelayVano Usage Page: Claude and GPT Costs on One API Relay

What the RelayVano Usage Page Shows for Every Request

When you run requests through RelayVano, you do not have to guess what a call cost. Our usage page lists every request with the tokens it consumed and the charge that was applied to your balance. Prompt tokens, reply tokens, cache stores and cache hits are priced separately, and each of them is visible in the record for that request. That makes it straightforward to see which part of a conversation is driving the spend, whether you are testing a prompt once or running a steady workload all week.

The billing model underneath is deliberately plain. You top up a prepaid balance and we charge per token; there is no monthly fee, and the balance never expires. Unused credit stays where it is until you spend it or request a refund inside the refund window described below. Since each request carries its own charge, your usage page doubles as a running statement of what your day of development actually cost.

One detail is worth knowing before you start: if a caller disconnects mid-stream, the request is still finished and billed on the tokens really used. Aborting a stream does not erase the work, so it is worth planning long replies with that in mind.

For a quick sense of what the page can show, a single line can carry a prompt count, a reply count, a cache store entry, a cache hit count and a final amount. Nothing is folded into an estimate you have to interpret later, and nothing arrives as one monthly summary with no detail behind it.

One Key, Three Protocols, and the Right Client for Each Model

RelayVano is an independent relay made for developers worldwide, and one API key reaches every model we serve through https://api.relayvano.com. The same key works across three protocol shapes: the OpenAI-compatible /v1 route, the Anthropic /v1/messages route, and the native Gemini /v1beta route. You pick the route that matches the client you already use, rather than learning a new toolchain to test a new model.

Client pairing is specific, and we keep it specific:

• 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, in Gemini CLI. They are not OpenAI-compatible, so an OpenAI-style client is not the right place to send them.

That last point saves setup time. If you are working in Cursor or Cline, choose from GPT, DeepSeek, MiniMax, Qwen, Kimi, Grok, Hunyuan or MiMo models. If your workflow is built around Claude Code or Claude Desktop, the Claude family is there for you. If you want Gemini, bring Gemini CLI and the native /v1beta route.

Because one key covers all of them, you can move a task between model families without rebuilding your credentials or your billing. The usage page follows you across whichever route you chose, so the numbers stay in one place as your experiments become a routine.

Token Fares on the Page and a 200-Request Example

Fares are shown as approximate US dollars per million tokens, with the actual charge made in yuan. A few examples: Claude Fable 5.1 is $4.07 prompt / $20.33 reply / $0.2222 cache hit, against an official list price of $10 / $50 / $0.25 cache hit. Claude Haiku 4.5 is $0.3958 prompt / $1.90 reply / $0.0319 cache hit, against $1 / $5 / $0.1 cache hit. GPT-6.1 Sol is $0.1958 prompt / $0.9792 reply / $0.0389 cache hit, against $2 / $10 / $0.1 cache hit. MiniMax M3 is $0.1472 prompt / $0.5889 reply / $0.0292 cache hit, and Gemini 3.8 Flash is $1.20 prompt / $5.40 reply / $0.1194 cache hit.

Here is a worked example, and it is only an example. Suppose you send 200 requests a day, each with 2,000 prompt tokens and 1,000 reply tokens on GPT-6.1 Sol. The prompt cost per request is $0.1958 × 2,000 ÷ 1,000,000 = $0.0003916. The reply cost per request is $0.9792 × 1,000 ÷ 1,000,000 = $0.0009792. That is $0.0013708 per request, or $0.27416 across 200 requests for the day. Your usage page shows those same token counts and the charge behind them, request by request, so the arithmetic is something you check rather than something you take on faith.

Cache behavior matters too. Cache stores and cache hits carry their own rates, so a workflow that reuses context can look different from one that resends it. On GPT-5.6 Sol, for instance, the fare is $0.3736 prompt / $2.09 reply / $0.0306 cache hit, compared with an official list price of $4 / $20 / $0.4 cache hit; that comparison reflects an OpenAI promotion stated to run at least until 2026-11-21.

Because the figures are per token and per request, small changes in prompt size show up as small, traceable amounts rather than a surprise on a single invoice. Claude Opus 5.5 at $1.65 prompt / $7.82 reply / $0.075 cache hit and Gemini 3.1 Pro at $1.58 prompt / $9.13 reply / $0.1569 cache hit sit in the same table, readable in the same way.

Top-ups by Stripe and Refunds Within 9 Days

Adding credit is handled through Stripe checkout. Cards work everywhere, and WeChat Pay is available where your account supports it. You choose the amount, confirm the payment, and the balance is ready for your next request. New accounts start with $0.40 of credit, which is enough to send a first request and see exactly how the usage page records it.

Unspent top-ups can be refunded within 9 days. The window is short enough to keep in mind if you are testing an unfamiliar model, and long enough to try one before deciding whether it belongs in your stack. We would rather you know the terms up front than discover them after a large top-up.

There is no subscription to cancel and no recurring charge to track. If usage is quiet for a week, nothing renews and nothing expires; the balance simply waits, and the usage page keeps its history so you can see which days were busy and which were not. For teams comparing options, note that RelayVano sells pay-as-you-go balance rather than a vendor's subscription, so the cost you can forecast is the cost of the tokens you actually send.

Because top-ups are one-off payments, you can scale them to your own pace: a small amount while you evaluate a model, a larger one once a workflow proves itself in your own usage records.

Six Languages, Referral Earnings, and Direct Support

RelayVano was made for developers worldwide, and that shows in the interface itself. The whole site, dashboard and docs come in six languages: 中文, English, Español, हिन्दी, Português, Français. Prices show in yuan on the Chinese pages and in approximate dollars on the others, so a developer reading the docs in any of those languages sees the same service and the same numbers in a familiar form.

Referrals are simple to explain. People who sign up through your link bring you 16% of every top-up they make, and you also earn 4% one level further. Earnings can be withdrawn from $5.56 or moved straight into your balance, where they fund your next request without a separate step.

If something does not look right, support is one click away: logged-in members write to us under Messages in the dashboard, where the request and its usage record are already in view. That makes it easier for us to understand the case, and easier for you to keep working while we look. When a charge on the usage page needs explaining, the tokens behind it are right there in the same conversation.

Start with the $0.40 credit, send a request through the client that matches your model, then open the usage page and read the tokens and charge for yourself. When you are ready to keep going, sign up for RelayVano and put your first top-up behind the API key you already know how to use.