One Large Code Review on RelayVano: Claude API Relay Pricing

Fare watch EN2026-11-03

What one large code review costs at today's fares

Code review feels like a small task until you hand a repository to a model. A single pass over a busy service can mean tens of thousands of prompt tokens, a long generated reply, and a large volume of cache hits from files the model has already seen. Multiply that by a team shipping every day, and the number on your bill starts to matter.

Here is a worked example, and we want to be clear that it is only an example. Suppose one large review sends 60,000 prompt tokens, receives 20,000 reply tokens, and records 300,000 cache-hit tokens. At RelayVano's current Claude Sonnet 5.5 fares of $1.25 per million prompt tokens, $6.25 per million reply tokens and $0.0889 per million cache-hit tokens, that single review comes to roughly $0.23. The prompt portion is about $0.075, the reply about $0.125, and the cache hits about $0.027.

Now scale it. If your team runs ten reviews of that size on an ordinary weekday, that is roughly $2.27 for the day, again as an example rather than a promise about your own workload. Cache stores are priced separately from cache hits, so the usage page shows each category on its own line and you can see exactly which part of the request produced the charge.

The useful thing about this calculation is that nothing is hidden behind a subscription. RelayVano bills per token against a prepaid balance, there is no monthly fee, and the balance never expires. You pay for what the model actually reads and writes.

Step 1: Choose the model and the client that matches it

Getting the client pairing right is the step that saves the most wasted tokens, because a mismatched setup either fails outright or sends you down a more expensive path than you intended.

Step 1a: If your review runs on Claude models, including Claude Sonnet 5.5, Claude Opus 5.5, Claude Haiku 4.5 and Claude Fable 5.1, use the official Claude Code or Claude Desktop clients, version 2.1.285 or newer. Claude models work only in those two clients.
Step 1b: If you prefer Gemini models such as Gemini 3.1 Pro, Gemini 3.8 Flash or Gemini 4 Argon, use the native Gemini protocol, which means Gemini CLI. Gemini models are not OpenAI-compatible, so an OpenAI-style client will not drive them.
Step 1c: If you live in Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs or cURL, pair those with GPT, DeepSeek and the other non-Claude, non-Gemini models. That includes options such as GPT-6.1 Sol at $0.1958 per million prompt tokens, DeepSeek V4 Flash at $0.3097, Qwen3.7 Plus at $0.1458, MiniMax M3 at $0.1472 and MiMo V2.6 Pro at $0.2319.

Notice what this means for a code review. A Claude-powered review belongs in Claude Code or Claude Desktop. A review you want to run inside Cursor or through cURL belongs on a GPT, DeepSeek, Qwen or similar model. Choosing deliberately lets you compare like with like, instead of comparing a Claude run against a completely different workflow.

Step 2: Point one key at every model

RelayVano is an independent relay made for developers worldwide, and the setup is deliberately small. One API key reaches every model through the same base address, https://api.relayvano.com. From there, the paths follow each model's own protocol: an OpenAI-compatible /v1 route, an Anthropic /v1/messages route, and a native Gemini /v1beta route.

In practice, you create a key once, set it as your base address in the client you picked in Step 1, and choose the model by name. There is no separate account to open for Claude, GPT or Gemini, and no per-vendor billing relationship to set up. The whole site, dashboard and documentation 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, with the charge itself made in yuan.

Because the same key covers every protocol, a team can split its review work across models without changing infrastructure. A deeper architectural pass might use Claude Opus 5.5, a fast triage pass might use Gemini 3.8 Flash, and a scripted check might run through cURL against a non-Claude model, all while the usage page records each request with its charge.

Step 3: Fund the balance, then read the usage page

RelayVano sells pay-as-you-go balance and nothing else. New accounts start with $0.40 of credit, which is enough to run a small review and see the whole flow work before you commit any larger amount.

Top-ups go through Stripe checkout, so cards work everywhere, and WeChat Pay is available where your account supports it. Unspent top-ups can be refunded within 9 days. There is no monthly fee to remember and no renewal date to track, because the balance never expires and you simply draw it down as requests arrive.

Once you are running, the usage page is where the real picture appears:

Every request appears with its charge, so a single review can be traced line by line.
Prompt tokens, reply tokens, cache stores and cache hits are priced separately, which is what lets you see whether a long context is worth its cost.
If a caller disconnects mid-stream, the request is still finished and billed on the tokens really used, so an interrupted client session does not silently hand you a partial bill.

If you need help with any of it, logged-in members write to us under Messages in the dashboard. That keeps the conversation next to the account, the balance and the request history.

Step 4: Compare a premium pass against a lighter one

The example review above used Claude Sonnet 5.5. Suppose you decide that a particularly delicate change deserves a stronger model, and you run the same 60,000 prompt tokens, 20,000 reply tokens and 300,000 cache-hit tokens through Claude Opus 5.5, at $1.65 per million prompt tokens, $7.82 per million reply tokens and $0.075 per million cache-hit tokens. That example comes to roughly $0.28, against roughly $0.23 for Sonnet 5.5.

For comparison, the official list price for Claude Sonnet 5.5 is $2 per million prompt tokens, $10 per million reply tokens and $0.1 per million cache-hit tokens, which puts the same example review at about $0.35 on list pricing. Other Claude fares are published the same way, including Claude Fable 5.1 at $4.07 prompt and $20.33 reply, Claude Opus 5 at $1.88 prompt and $9.38 reply, and Claude Haiku 4.5 at $0.3958 prompt and $1.90 reply, so you can weigh a faster model against a pricier one before you spend.

RelayVano also shares a small part of the top-up when you bring colleagues along. People who sign up through your link generate 16% of every top-up for you, and 4% one level further. Earnings can be moved straight into your balance or withdrawn from $5.56, which is convenient if you would rather spend the credit on the next review than cash it out.

Step 5: Sign up and price your own review

Here is the short version of the walkthrough. Pick the model that fits your client, point one key at https://api.relayvano.com, add credit through Stripe checkout, and let the usage page show you what each request cost. The example review priced above lands at roughly $0.23 on Claude Sonnet 5.5 and roughly $0.28 on Claude Opus 5.5, with no subscription in the way and no balance expiry to plan around.

New accounts start with $0.40 of credit, so you can test the whole path today rather than theorize about it. RelayVano is an independent relay made for developers worldwide, with a dashboard and documentation in six languages, per-token billing you can inspect request by request, and support sitting under Messages whenever you need it. Sign up, run one large review, and read the charge on the usage page for yourself.