Claude Sonnet 5.5 API relay on Vana API: $1.18 per million input

§ One address, three protocols, minimal setup

Most of the effort in a pay-as-you-go setup happens before a token is billed. You locate the endpoint, create a key, and point the right client at the right protocol. Vana API keeps that short. One address, https://api.vanaapi.com, answers API requests, while the dashboard for keys, balance and usage log sits on the main site, https://vanaapi.com. There is no monthly fee, and the operational pieces stay together instead of being spread across separate services.

Three protocols share that single address. Claude models speak Anthropic Messages (/v1/messages) and are used only from official Claude Code or Claude Desktop, version 2.1.285 or later, which speak that protocol themselves. Gemini models use the Gemini native protocol (/v1beta/models/...) from Gemini CLI. The GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo families use the OpenAI format (/v1/chat/completions), and GPT models also answer on /v1/responses.

For a team that only needs Claude Sonnet 5.5, the setup reduces to the Claude side of that arrangement: a supported client version, a key created in the dashboard, and one address to call. The wider protocol coverage matters when you add other families later, and adding them does not mean moving to a different host or opening a second account. That is the shape of the service: one address, three protocols, minimal setup.

§ The unit prices behind the sample week

Claude Sonnet 5.5, model id claude-sonnet-5-5, is priced on Vana API at $1.18 per million input tokens and $5.76 per million output tokens. The official list price for the same model is $2 / $10. Every figure used in this article starts from those two published rows, and every total below is the result of multiplying a real token count by those rates. Nothing is invented, and no rate is adjusted in either direction.

Output tokens cost more than input tokens on Claude Sonnet 5.5, so the shape of a week depends on how much text comes back as well as how much goes in. That is why the example below is written as an input and output pair rather than a single blended token figure. Two teams sending the same number of messages can spend different amounts over the same seven days simply because their responses differ in length.

Two other Claude models are worth noting for the same workflow. Claude Haiku 4.5, model id claude-haiku-4-5-20251001, is listed on Vana API at $0.3472 / $1.60 per million tokens, against an official list price of $1 / $5. Claude Opus 5.5, model id claude-opus-5-5, is listed at $1.46 / $6.94, against an official list price of $4 / $20. If part of your week is short and routine, routing those requests to Claude Haiku 4.5 while keeping Claude Sonnet 5.5 for the rest is a direct way to shape the total. Both are reached the same way, over the same Anthropic Messages path on the same Vana API address.

§ A sample week of Claude Sonnet 5.5, worked out

Here is an example. Say you run this every day: 1,000,000 input tokens and 1,000,000 output tokens, across seven equal working days. On Vana API, the input side is 1,000,000 tokens at $1.18 per million, so $1.18 per day. The output side is 1,000,000 tokens at $5.76 per million, so $5.76 per day. Together that is $1.18 + $5.76 = $6.94 per day, and the sample week is 7 × $6.94, made up of 7 × $1.18 on input and 7 × $5.76 on output.

Run the same example against the official list price of $2 / $10 for Claude Sonnet 5.5. The input side becomes $2 per day and the output side becomes $10 per day, so $2 + $10 = $12 per day and 7 × $12 across the week. The gap between the two is 7 × ($12 − $6.94), which follows directly from the two published rate pairs. No other assumption enters the calculation: the same token counts, the same seven days, two different rate cards.

A lighter pattern is worth checking. Here is another example. Say you run this every day: 1,000,000 input tokens and 1,000,000 output tokens, but only on three days a week. On Vana API, that is 3 × $1.18 on input and 3 × $5.76 on output, for a weekly total of 3 × $6.94. At the official list price of $2 / $10, the same pattern gives 3 × $2 plus 3 × $10, or 3 × $12, and the difference is 3 × ($12 − $6.94). A five-day pattern would use the same two rates with 5 in place of 3.

These are examples, not quotas or forecasts. What they show is the method: take your real input and output token counts, multiply them by $1.18 and $5.76 per million on Vana API, and you have the week before you commit to it. If your actual usage differs, the arithmetic differs with it. And because every call is listed in the usage log in the dashboard, the worked-out figure can be checked against recorded entries afterwards rather than trusted on faith.

§ Prepaid balance, monthly plans and the practical terms

Vana API works on a prepaid balance, charged per real token, with no monthly fee. That fits a pay-as-you-go budget well: you fund the balance, and calls draw it down as they happen. Keys are created in the dashboard, and every call is listed in the usage log there, so the sample-week arithmetic above can be reconciled against actual token counts and actual charges.

There is a gift of $0.10 of credit on sign-up. That is enough to make a small test call and confirm that the endpoint, the key and the client version are all in the right order before a larger top-up.

If you would rather buy a fixed allowance, two monthly plans cover Claude models only. One is Claude up to $450 / 30 days for $64.72 for 30 days, at most $140 a week. The other is Claude up to $1,050 / 30 days for $153.61 for 30 days, at most $320 a week. A team running a steady pattern like the seven-day example above can pick the allowance that matches its measured volume instead of topping up repeatedly, and the weekly limits make the comparison plain.

Refunds are handled directly as well. Unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee. The referral programme pays 14% of what people you invite pay, and 5% one level further, with earnings that can be withdrawn from $13.89.

§ Setting up Claude Sonnet 5.5 and reading the log

Putting Claude Sonnet 5.5 on Vana API takes a short sequence. Start on the dashboard at https://vanaapi.com and create a key there. Open official Claude Code or Claude Desktop and confirm the version is 2.1.285 or later, since that is the range that speaks Anthropic Messages (/v1/messages) itself. Then point it at https://api.vanaapi.com, the address that answers API requests, and use the model id claude-sonnet-5-5.

Send a small request and open the usage log in the dashboard. You will see the call, the tokens it consumed and the amount taken from the prepaid balance. From that point, the weekly figure is the sum of the entries you recorded, and it should match the $1.18 and $5.76 per million rates once you multiply them by the logged token counts.

Two details keep the running cost predictable. Output tokens are billed at $5.76 per million while input tokens are billed at $1.18 per million, so keeping prompts and context lean matters more than trimming replies. And the monthly Claude plans are measured in a 30-day period with a stated weekly limit, which makes each one easy to compare against a week you have already measured on Vana API.

That is the whole arrangement. Vana API holds the supported families behind one address, keeps keys, balance and usage in one dashboard, and bills on real tokens through a prepaid balance with no monthly fee. For Claude Sonnet 5.5, the published rates of $1.18 and $5.76 per million give you a number you can multiply before you commit, and the usage log gives you the actual figure afterwards. One address, three protocols, minimal setup, and a weekly cost you can work out from prices that are stated openly.