Claude Haiku 4.5 Pricing on the RelayVano API Relay

Why Claude Haiku 4.5 works well as a first reader in Claude Code

Most Claude Code sessions start the same way: you paste in a repository, a ticket thread, a stack trace or a long document, and wait for a model to read all of it before saying anything useful. That first read is often the most expensive part of the loop, because the volume of input is large even though the task is simple. Sorting files, locating a function, summarizing a diff or skimming an error log does not usually need the heaviest model available.

Claude Haiku 4.5 is a good fit for that role. It is a public Anthropic model, and the point of this article is not that one model replaces the others but that a fast, inexpensive first reader changes how the rest of your session feels. Read first with Haiku, escalate only the part that needs deeper reasoning, and your token spend becomes a deliberate choice instead of a side effect.

RelayVano is an independent relay made for developers worldwide, and it is built so this setup stays simple. One API key reaches every model through https://api.relayvano.com, and 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 in any of those languages can read the same numbers and understand the same setup.

Where Claude Code setups usually go wrong

The most common failure has nothing to do with the model. It happens because a configuration written for one family gets pasted into another. On RelayVano, the client pairing is defined clearly. Claude models work only in the official Claude Code and Claude Desktop clients, version 2.1.285 or newer. If you try to drive Claude Haiku 4.5 from Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs or cURL, the request will not work, because those clients pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models.

The second mistake is sending the request down the wrong path. Behind the single RelayVano key there are three routes: an OpenAI-compatible /v1 path for the non-Claude, non-Gemini models, an Anthropic /v1/messages path for Claude, and the native Gemini /v1beta path for Gemini models. Gemini models use the native Gemini protocol only, through Gemini CLI; they are not OpenAI-compatible. Mixing the Anthropic route with an OpenAI-style payload is a frequent reason a first attempt returns an error that looks like a bad key.

The third mistake is assuming one flat price per message. Prompt tokens, reply tokens, cache stores and cache hits are each priced separately. A chatty assistant that replies at length can cost several times more than a terse one even when the input is identical, and a long context that gets cached repeatedly behaves differently again. RelayVano lists every request with its charge on the usage page, so you can see which side of the exchange the tokens went to.

The fourth mistake is treating an interrupted stream as a free request. If a caller disconnects mid-stream, RelayVano still finishes the request and bills it on the tokens really used. That is worth knowing before you build retry loops around a flaky network, because a retry is a second billable request, not a cancelled one.

Claude Haiku 4.5 pricing and a 200-request example

Here is what the fast first reader costs on RelayVano, in approximate US dollars per million tokens, with the charge itself made in yuan: Claude Haiku 4.5 is $0.3958 prompt, $1.90 reply and $0.0319 cache hit. The official list price for comparison is $1 / $5 / $0.1 cache hit. Cache hits are where a repeated context earns its keep, and at $0.0319 per million they are priced well below both prompt and reply rates.

To make the numbers concrete, consider this example. Say you send 200 requests a day, and each one carries 1,000 prompt tokens and 500 reply tokens. That works out to 200,000 prompt tokens and 100,000 reply tokens per day. Working from the fares above, the prompt side costs about $0.0792 and the reply side about $0.19, for a total of roughly $0.27 per day and about $8.07 over 30 days. This is an example only, not a forecast of your own usage, and it leaves out cache stores and cache hits entirely. Add those and the shape of the bill changes, which is exactly why the usage page matters.

Compare that with the heavier options in the same catalogue: Claude Sonnet 5 is listed at $1.27 prompt and $6.36 reply per million tokens, and Claude Opus 5 at $1.88 prompt and $9.38 reply. The gap is large enough that the order in which you call models matters. Let Claude Haiku 4.5 read and summarise, then bring in a stronger model for the reasoning that actually needs it. RelayVano puts all of them behind the same key and the same balance, so you can switch without changing your account.

Setting up Claude Code with one RelayVano key

The setup takes a few minutes and does not require a monthly fee or a vendor's subscription. First, create your RelayVano account. New accounts start with $0.40 of credit, which is enough to send a test request and confirm that the whole path works before you add any money.

Next, top up through Stripe checkout. Cards work everywhere, and WeChat Pay is available where your account supports it. Unspent top-ups can be refunded within 9 days. RelayVano sells nothing but a pay-as-you-go balance: you pay per token, there is no monthly fee, and the balance never expires, so an unused credit is not a deadline you have to beat.

Then point Claude Code at RelayVano. In your client settings, set the Anthropic base URL to https://api.relayvano.com and route requests through the Anthropic /v1/messages path. Put your RelayVano key into the environment variable your Claude Code installation reads for the Anthropic key, and check that your client is version 2.1.285 or newer. If you work at a desk rather than in a terminal, Claude Desktop accepts the same Claude models and the same route, so the same key covers both.

Finally, send one short request and open the usage page. You should see the request listed with its charge. If nothing appears, the usual causes are an outdated client or a mismatched route rather than anything wrong with the model itself. If you get stuck, logged-in members can write to us under Messages in the dashboard, where the conversation stays attached to your account.

One key, many models, six languages

Once the Claude route is working, the rest of the catalogue opens up without new credentials. The same key reaches GPT, DeepSeek and the other non-Claude, non-Gemini models through the OpenAI-compatible /v1 path, which is what Codex, Cursor, Cherry Studio, Cline, OpenCode, the OpenAI SDKs and cURL pair with. Gemini models come through the native Gemini /v1beta path in Gemini CLI. One account, one balance, three routes, and a client pairing that is documented rather than guessed.

That matters because a relay is only useful if it stays predictable as your habits change. Today you may want Claude Haiku 4.5 to scan a codebase before Claude Sonnet 5.5 takes over the design work. Tomorrow you may want GPT-6.1 Sol in Cursor, or DeepSeek V4 Pro for a long batch job. Because billing is prepaid balance with per-token pricing and the balance never expires, switching models is a configuration choice rather than a purchase decision.

There is also a way to share the value. Referrals pay 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 your balance, so a referral can fund the next month of usage without another checkout.

RelayVano is an independent relay made for developers worldwide, available in six languages with transparent per-request charges and no expiring credit. Sign up for RelayVano, add a small top-up, and make Claude Haiku 4.5 your first reader in Claude Code.