Vana API relay for Claude Fable 5.1: $3.33 per million input tokens

§ One address, three protocols, minimal setup

Vana API puts three protocol families behind a single address, https://api.vanaapi.com. Claude models arrive over Anthropic Messages (/v1/messages). Gemini models arrive over the Gemini native protocol (/v1beta/models/...). GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo arrive in the OpenAI format (/v1/chat/completions), and GPT models also answer on /v1/responses. One base URL, three protocols, and a small amount of setup before a team starts sending real work.

For a company, the practical gain is that the base configuration stops being the project. There is one address to record, and it stays the same when a team changes model family. The dashboard remains at https://vanaapi.com, where keys are created, balance is held, and the usage log is kept. The API address answers API requests; it does not take the place of the dashboard. Nothing about the routing changes between a first test call and a production workload, because there is only one route to remember.

Minimal setup also means the decision is made once per model rather than once per team. Claude models are used from official Claude Code or Claude Desktop, version 2.1.285 or later, since those clients speak Anthropic Messages themselves. Claude Fable 5.1 is no different in this respect: it is reached from official Claude Code or Claude Desktop, version 2.1.285 or later, and nowhere else. The model choice determines the client. The address does not change.

§ What Claude Fable 5.1 costs per million tokens

Claude Fable 5.1, model id claude-fable-5-1, is the model this example follows. Its Vana API price is $3.33 per million input tokens and $16.74 per million output tokens, against an official list price of $10 and $50. That difference is the reason to look at a relay service in the first place, and the reason to check the arithmetic rather than accept a headline on faith.

The other Claude models on the sheet are priced the same way, per real token, with no monthly fee attached to the rate. Claude Haiku 4.5 (claude-haiku-4-5-20251001) is $0.3472 input and $1.60 output per million tokens, against a list price of $1 and $5. Claude Sonnet 5.5 (claude-sonnet-5-5) is $1.18 and $5.76, against $2 and $10. Claude Opus 5.5 (claude-opus-5-5) is $1.46 and $6.94, against $4 and $20. A team can move between them on the same address, with the same key, and see each call itemised in the usage log.

One condition applies to all of them: Claude models are used from official Claude Code or Claude Desktop, version 2.1.285 or later. That is not a preference. It follows from how the protocol is spoken, and it keeps the setup short, because the client already knows the shape of the request that Claude Fable 5.1 expects.

§ Worked example: a month of Claude Fable 5.1

Here is an example, not a customer result. Say a company runs Claude Fable 5.1 through Vana API every working day and, across 30 days, sends 10,000,000 input tokens and 2,000,000 output tokens. Every figure below is an example calculation built from the published per-million prices for Claude Fable 5.1.

Input: 10,000,000 tokens is 10 million tokens, and 10 × $3.33 = $33.30. Output: 2,000,000 tokens is 2 million tokens, and 2 × $16.74 = $33.48. Total for the 30-day example: $33.30 + $33.48 = $66.78. Spread across the same 30 days, that is $66.78 ÷ 30 = $2.226 per day, and $66.78 ÷ 30 × 7 = $15.582 per week.

The same example at official list prices: 10 × $10 = $100 for input, 2 × $50 = $100 for output, so $200 for the 30 days. The difference between $200 and $66.78 is $133.22 for the example month. The example week of $15.582 sits inside the $140 a week ceiling on the smaller Claude monthly plan, though the $64.72 price of that 30-day plan would not cover a $66.78 month. The larger Claude plan, $153.61 for 30 days up to $1,050, with at most $320 a week, would cover the example.

Because Vana API charges per real token against a prepaid balance, with no monthly fee, the company can also skip a plan entirely and fund $66.78 directly. Either way, every call in the example appears in the usage log in the dashboard, so the $66.78 can be traced back to the calls that produced it. That traceability matters more than a headline rate: it lets a finance team reconcile a month of Claude Fable 5.1 usage against a single line item.

§ Prepaid balance, monthly plans, and the numbers around them

Money in is prepaid. Balance is drawn down per real token, and there is no monthly fee on top. The gift on sign-up is $0.10 of credit, which is small enough to treat as a starting test rather than a budget line. Unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee, so the top-up size is worth choosing with that window in mind.

The monthly plans are stated as caps rather than token allowances. For Claude models only, the first plan covers up to $450 per 30 days for $64.72, with at most $140 a week; the second covers up to $1,050 per 30 days for $153.61, with at most $320 a week. For GPT models only, the plan covers up to $2,750 per 30 days for $103.61, with at most $820 a week. A plan that is not the right family for the model in use is not a saving, so the family label is the first thing to check when a Claude Fable 5.1 workload is being planned.

Referrals are also priced in plain percentages. You receive 14% of what people you invite pay, and 5% one level further. Earnings can be withdrawn from $13.89. That threshold is worth noting before treating referral income as working capital, and it is stated in the same terms as everything else in the pricing sheet.

§ What setup looks like for a team

Setup, in full: create a key in the dashboard at https://vanaapi.com, add credit, and point official Claude Code or Claude Desktop, version 2.1.285 or later, at https://api.vanaapi.com with the model id claude-fable-5-1. Then watch the usage log in the dashboard, where every call is listed and the balance moves. There is no second address to carry into procurement documents or internal documentation, and no per-model endpoint to maintain.

One address, three protocols, and a setup a team can describe in a sentence. What remains is a normal operating question: which model, how many tokens, and what the balance looks like afterwards. Vana API answers those with per-token prices, a prepaid balance that stays visible, and a usage log that records every call.

For a company weighing a relay service, the example above is the shape of the decision. Take the model, take the token volume, multiply by the per-million price, and compare the total against the list price. The arithmetic is short, the setup is short, and the numbers stay the same for everyone who reads them.