RelayVano API Relay Pricing: Claude Prompt, Reply and Cache Costs

Fare watch EN2026-11-02

Four charges, one usage line: how RelayVano bills a request

Open the usage page on RelayVano and you will find that a single request can produce more than one kind of charge. Prompt tokens, reply tokens, cache stores and cache hits are all priced separately, and every request appears with its charge attached. That detail matters more than it first sounds, because a line item that looks like one number on a provider's invoice can actually be four different fares moving in four different directions.

Prompt tokens are the input: the instructions, the code, the conversation history you send. Reply tokens are the output: what the model writes back. A cache store is the cost of writing a reusable block of context into the cache. A cache hit is the cost of reading that context back later. If you send the same large context repeatedly, caching can shift a meaningful share of your traffic from the prompt fare to the cache-hit fare, which is why the cache number deserves a place in your decision checklist rather than an afterthought.

The billing model itself is straightforward. RelayVano runs on a prepaid balance and pay per token, with no monthly fee, and the balance never expires. There is no vendor subscription to renew and nothing sitting idle behind a recurring charge. If a caller disconnects mid-stream, the request is still finished and billed on the tokens really used, so a dropped connection does not turn into a mystery line on your statement. You can check how the money moved on the usage page, request by request.

Client pairing: Claude Code, Gemini CLI and everything else

Before you compare a single fare, confirm that the client you actually use can talk to the model you want. One API key reaches every model through https://api.relayvano.com, using an OpenAI-compatible /v1 path, an Anthropic /v1/messages path and a native Gemini /v1beta path. The pairing, however, is deliberate and strict, and getting it wrong wastes a setup afternoon.

Claude models work only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. No other client is a supported route for them, so settle this point before you plan your workflow.
Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other models outside the Claude and Gemini groups. This is the natural path for many existing developer setups.
Gemini models use the native Gemini protocol only, through Gemini CLI. They are not OpenAI-compatible, so an OpenAI-shaped client configuration is the wrong route for them.

That split is the practical first question in any signup decision: does the tool you have open match the model you want to call? Answer it once, and the rest of the comparison becomes arithmetic instead of troubleshooting.

Comparing the prompt, reply and cache-hit fares you will actually see

The fares below are approximate US dollars per million tokens, with 1 USD ≈ ¥7.2, while the charge itself is made in yuan. On the Chinese pages prices are shown in yuan; on the other language pages they appear in approximate dollars. RelayVano is an independent relay made for developers worldwide, and the whole site, dashboard and docs come in six languages: 中文, English, Español, हिन्दी, Português, Français.

Among the Claude family, Claude Sonnet 5.5 is listed at $1.25 prompt, $6.25 reply and $0.0889 cache hit, while Claude Opus 5.5 sits at $1.65 prompt, $7.82 reply and $0.075 cache hit. Claude Haiku 4.5 comes in at $0.3958 prompt, $1.90 reply and $0.0319 cache hit, and Claude Opus 5 at $1.88 prompt, $9.38 reply and $0.1222 cache hit. For comparison, the official list prices quoted for those same models are $2 / $10 / $0.1 for Claude Sonnet 5.5, $4 / $20 / $0.2 for Claude Opus 5.5 and $1 / $5 / $0.1 for Claude Haiku 4.5.

On the GPT side, GPT-6.1 Sol is $0.1958 prompt, $0.9792 reply and $0.0389 cache hit, and GPT-6 Astra is $0.7403 prompt, $3.70 reply and $0.0736 cache hit. GPT-5.6 Sol is $0.3736 prompt, $2.09 reply and $0.0306 cache hit, against an official list price of $4 / $20 / $0.4 cache hit, with that OpenAI promotion stated to run at least until 2026-11-21. The Gemini entries include Gemini 3.8 Flash at $1.20 prompt, $5.40 reply and $0.1194 cache hit, and Gemini 3.1 Pro at $1.58 prompt, $9.13 reply and $0.1569 cache hit.

Several models outside those two groups are priced lower again at the entry point. MiniMax M3 is $0.1472 prompt, $0.5889 reply and $0.0292 cache hit. DeepSeek V4 Flash is $0.3097 prompt, $1.24 reply and $0.008333 cache hit, and MiMo V2.6 Pro is $0.2319 prompt, $0.4653 reply and $0.002778 cache hit. Rather than reading a long table end to end, pick the two or three models your work actually needs and compare their prompt, reply and cache-hit fares side by side.

Payments, refunds and the account you start with

Top-ups run through Stripe checkout, where cards work everywhere, plus WeChat Pay where your account supports it. Unspent top-ups can be refunded within 9 days, which makes a small first purchase a low-commitment way to test your setup. New accounts start with $0.40 of credit, enough to send a few test requests and confirm that your client, protocol and model are paired correctly before you commit to a larger balance.

RelayVano also runs a referral scheme: 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 straight into the balance, so a referral can either pay out or reduce what you spend. If anything is unclear, logged-in members can write to us under Messages in the dashboard, where the conversation sits next to the account you are asking about.

Because the balance never expires and there is no monthly fee, the question is not whether you will use the money in a particular month. It is simply how much you want to have available when a request arrives.

A worked example: 200 requests a day

Here is an example, not a forecast. Suppose you send 200 requests a day, each carrying 1,000 prompt tokens and receiving 500 reply tokens. That is 200,000 prompt tokens and 100,000 reply tokens per day. Using Claude Sonnet 5.5 fares, the prompt side costs about $0.25 a day, the reply side about $0.625, and the whole example adds up to roughly $0.875 a day, or about $26.25 across 30 days.

Run the same example against MiniMax M3 and the daily figure falls to roughly $0.088, or about $2.65 across 30 days. Now add caching to the picture: if part of that 200,000 prompt tokens is served from the cache, the cached portion is charged at the cache-hit fare instead of the prompt fare, and the daily total moves down again. The usage page shows which charge applied to each request, so your estimate can be checked against real numbers rather than assumptions.

A short signup checklist for RelayVano

Before you create an account, walk this list once. It takes a minute and saves a lot of guesswork later.

Decide which client you will drive: Claude Code or Claude Desktop for Claude models, Gemini CLI for Gemini models, or one of Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL for GPT, DeepSeek and the remaining models.
Choose the model, then read its three fares together: prompt, reply and cache hit. A cheap prompt with an expensive reply can cost more in practice than the other way round.
Estimate your daily token volume, and remember that cached context is billed differently from fresh context.
Top up a small amount through Stripe checkout, using a card or WeChat Pay where your account supports it, and note the 9-day refund window for unspent top-ups.
Check the usage page after your first requests, and use the $0.40 starter credit to confirm that every charge looks the way you expected.
Pick your dashboard language from the six available, and remember that logged-in members can reach us under Messages.

RelayVano keeps the moving parts small: one key, one prepaid balance, one usage page that shows what each request cost, and a client pairing that is documented rather than guessed. Start with the $0.40 credit, test the pairing, read the four fares on your first invoice line, and then top up only what your own usage justifies. Sign up when you are ready, and let the usage page do the talking.