Vana API setup for Claude Code: one address, $0.10

§ One address, three protocols

Vana API is built around a simple idea: a small team should be able to point its tools at one place and start working, without a long checklist of keys, accounts and endpoints. That place is https://api.vanaapi.com. It answers API requests, and that is all it does. Keys, balance and the usage log are managed separately, in the dashboard on the Vana API site at vanaapi.com. Separating the two keeps the request path clean and the account view clear.

The reason setup is short is that one address speaks three protocols. Anthropic Messages, the Gemini native protocol and the OpenAI chat completions format all arrive at the same base URL. You do not choose a different relay service per model family, and you do not maintain a list of gateways. You choose the client that matches your model, enter the Vana API address and your key, and the protocol does the rest.

Most guides begin with account creation, then an approval step, then a documentation hunt. With Vana API, the practical sequence is shorter: create a key in the dashboard, top up the prepaid balance, and enter the address in the client you already use. For a team of a few people, that is most of the work, and it is work you can hand to someone else without a handover document.

§ Claude Code and Claude Desktop

Claude models are reached through the official Claude Code or Claude Desktop applications, version 2.1.285 or later. Those applications already speak Anthropic Messages (/v1/messages), so pointing them at Vana API is a configuration change rather than an integration project. Claude models are limited to those two applications, which means there is one path to get right and no second route to remember when something needs checking.

For a small team working with Claude Code, that path is short. Create a key in the dashboard, set the base address to https://api.vanaapi.com, and choose the Claude model you want. Requests travel in the format the application already expects, and each one is charged per real token. There is no monthly fee on the balance you top up, and no separate subscription to keep track of before the first call.

It is worth being precise about what this setup covers. Because Claude models are served only to those two official applications, a teammate who opens a different editor and pastes the same address will not get an answer. That is a design choice rather than a gap to work around: the protocol handling stays inside the applications, and Vana API does not attempt to translate it into something else.

For teams, the practical benefit is that onboarding a new person does not depend on model-specific knowledge. The address, the key and the model name are the same three pieces of information for everyone, and the dashboard shows what each person has been spending.

§ Gemini CLI and the Gemini native protocol

Gemini models use the Gemini native protocol (/v1beta/models/...), and Gemini CLI is the client that speaks it. The Gemini native protocol is not OpenAI-compatible, so a request shaped for one format will not be read by the other. If Gemini CLI is already part of your team's workflow, the Vana API setup follows the same pattern: one address, one key, one model name.

The reason this matters for a small team is that different people often use different tools. One developer may live in Claude Code, another may reach for Gemini CLI, and both can work from the same Vana API account, the same prepaid balance and the same usage log. You are not running two services. You are running one address with three protocols behind it, and one dashboard where the numbers are visible.

§ OpenAI-format models in everyday editors

GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo models use the OpenAI format (/v1/chat/completions). This is the format most editors and libraries already expect, so Cursor, Cline, Roo Code, OpenCode, Cherry Studio, any OpenAI SDK or cURL can all be pointed at the same address with the same key. GPT models also answer on /v1/responses, which is what Codex uses.

For teams, this is often the part that removes friction. Instead of configuring a separate provider for every tool, a shared address and a shared key can be handed around, and the usage log in the dashboard shows what each call actually cost. When you add a model or open another editor, the setup keeps the same shape, which is a useful property when the team is small and nobody wants to own a complicated configuration.

Because the same prepaid balance covers all three protocols, a team can test a different model family for one afternoon without opening a second billing relationship. The spending shows up in the same log, in the same dashboard, next to everything else.

§ What Claude models cost on Vana API

Prices are per million tokens, input and output. For the Claude models available through Vana API:

• Claude Fable 5.1 (claude-fable-5-1): $3.33 / $16.74, against an official list price of $10 / $50
• Claude Haiku 4.5 (claude-haiku-4-5-20251001): $0.3472 / $1.60, against an official list price of $1 / $5
• Claude Opus 5.5 (claude-opus-5-5): $1.46 / $6.94, against an official list price of $4 / $20
• Claude Sonnet 5.5 (claude-sonnet-5-5): $1.18 / $5.76, against an official list price of $2 / $10

To make the unit concrete, say a developer sends 1 million input tokens and 100,000 output tokens through Sonnet 5.5 in a week. The input side is $1.18. The output side is $0.576, because $5.76 per million output tokens is $0.576 per 100,000. That is the whole calculation: real tokens, real prices, and a log entry for each call.

Haiku 4.5 is priced at $0.3472 / $1.60, which suits short, frequent tasks, while Opus 5.5 at $1.46 / $6.94 is the higher-priced option for heavier work. Fable 5.1 sits between them at $3.33 / $16.74. The choice is made per call, and the usage log shows what you spent.

§ Prepaid balance, plans and refunds

Vana API bills from a prepaid balance, charged per real token, with no monthly fee. Keys are created in the dashboard, and every call is listed in the usage log there, so the numbers can be checked against the balance. New accounts receive a $0.10 credit gift on sign-up, which is enough to make a first call and see the log entry appear.

If usage is steady, monthly plans are also available:

• Claude up to $1,000 / 30 days: $64.72 for 30 days, at most $320 a week, Claude models only
• Claude up to $2,400 / 30 days: $153.61 for 30 days, at most $750 a week, Claude models only
• GPT up to $4,000 / 30 days: $103.61 for 30 days, at most $1,200 a week, GPT models only

Plans sit alongside the pay-as-you-go balance rather than replacing it, so a team can cover predictable work with a plan and keep using prepaid credit for everything else. Refunds are straightforward: unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee.

There is also a referral arrangement: 14% of what people you invite pay, and 5% one level further, with earnings withdrawable from $13.89. For a small team that already uses Vana API, inviting a colleague or a partner is a reasonable way to recover part of the spend.

Put together, the picture is simple to explain to a teammate: one address, three protocols, minimal setup, and a usage log that shows what each call cost. Everything else follows from those four points.