GPT-6 Astra pay-as-you-go pricing on Vana API: $103.61 for 30 days

§ One address, three protocols

If you are new to model APIs, Vana API is built around a simple idea: there is one address to learn and three protocol families to understand. The request address is https://api.vanaapi.com. That address answers API requests. Your keys, your balance and your usage log are on the main site, https://vanaapi.com, so the working surface and the account surface stay separate and easy to find.

Everything you need for a first call is short: create a key in the dashboard, add credit, point your tool at the one address, and choose the model id you want. There is no monthly fee on the pay-as-you-go side. You are charged per real token, and every call appears in the usage log, so you can see what each request cost.

The only decision left is which protocol your tool already speaks. Vana API maps the model families onto the protocol each tool expects:

• Claude models: only from official Claude Code or Claude Desktop, version 2.1.285 or later, which speak Anthropic Messages (/v1/messages) themselves. They do not work from any other client.
• Gemini models: only from Gemini CLI, over the Gemini native protocol (/v1beta/models/...). They are not OpenAI-compatible.
• GPT, Grok, DeepSeek, GLM, Kimi, Qwen, MiniMax and MiMo models use the OpenAI format (/v1/chat/completions). GPT models also answer on /v1/responses, which is what Codex uses.

Those Claude models are reachable only through the two applications named above. An SDK, cURL or a script cannot reach them, and no other client can either. That is why the version note matters: on Claude Code or Claude Desktop 2.1.285 or later, you can send Anthropic Messages requests straight to the one address.

For the OpenAI-format families, the same address works from Cursor, Cline, Roo Code, OpenCode, Cherry Studio, any OpenAI SDK or cURL. Gemini stays inside Gemini CLI, over the Gemini native protocol, and it is not OpenAI-compatible. The result is a setup with very little to configure: one address, three protocols, and a model id that tells Vana API what you want.

§ GPT-6 Astra and the GPT monthly plan

GPT-6 Astra (model id gpt-6-astra) is the model to plan around when a week gets heavy. On Vana API, GPT-6 Astra costs $0.6944 per million input tokens and $3.54 per million output tokens. The official list price for the same token counts is $10 / $50, so the Vana API rate is the number to put next to your budget for a busy week.

For sustained GPT work, the monthly option is the GPT plan: GPT up to $4,000 / 30 days for $103.61 for 30 days, at most $1,200 a week, GPT models only. The caps are real and stated up front. The plan covers a 30-day window, the weekly ceiling is $1,200, and the model scope is limited to GPT models, which keeps the plan focused on the family you are actually running.

Say you run a heavy week of GPT-6 Astra, with several days of long context and long outputs. The number to keep in view is $1,200 a week: that is the stated maximum for this plan, not an estimate. When a week approaches that ceiling, you know where you stand without reading a single invoice line, because each call is already itemised in the usage log on the dashboard.

Two lighter GPT models can sit alongside it when the task is smaller. GPT-5.6 Sol (model id gpt-5.6-sol) is $0.3472 / $1.81 per million tokens, against an official list price of $4 / $20. GPT-6.1 Sol (model id gpt-6.1-sol) is $0.1389 / $0.8333, against an official list price of $2 / $10. Both are GPT models, so both answer on the OpenAI format and on /v1/responses, which is what Codex uses.

§ Claude monthly plans and their weekly caps

When the work moves to Claude, Vana API offers two monthly plans, both scoped to Claude models only.

• 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.

The smaller plan suits a lighter Claude week, with a $320 weekly ceiling. The larger plan raises that ceiling to $750 a week and the 30-day limit to $2,400. Because both plans are Claude-only, a Claude-heavy week does not compete with GPT usage inside the same allowance, and a GPT-heavy week does not eat into a Claude plan.

The per-token rates for the Claude models on the list are set out clearly. Claude Fable 5.1 (model id claude-fable-5-1) is $3.33 / $16.74 on Vana API, against an official list price of $10 / $50. Claude Opus 5.5 (model id claude-opus-5-5) is $1.46 / $6.94, against $4 / $20. Claude Sonnet 5.5 (model id claude-sonnet-5-5) is $1.18 / $5.76, against $2 / $10. Claude Haiku 4.5 (model id claude-haiku-4-5-20251001) is $0.3472 / $1.60, against $1 / $5.

Keep the client rule in mind when you plan a Claude week: these models are reachable only from official Claude Code or Claude Desktop, version 2.1.285 or later. Set up the plan, then run the week from the application that already speaks Anthropic Messages.

§ Gemini, Grok, DeepSeek and the rest

Beyond GPT and Claude, the same one address carries more families, and the pay-as-you-go rates are per million tokens, input first and output second. Gemini 2.5 Flash (model id gemini-2.5-flash) is $0.2083 / $1.60. Gemini 3.1 Pro (model id gemini-3.1-pro) is $1.39 / $7.92. Gemini 3.8 Flash (model id gemini-3.8-flash) is $1.04 / $4.65. Gemini 4 Argon (model id gemini-4-argon) is $1.46 / $6.88.

Those models are used only from Gemini CLI over the Gemini native protocol, and they are not OpenAI-compatible. The OpenAI-format families fill out the rest of the catalogue. Grok 4.7 (model id grok-4.7) is $0.4861 / $1.53. DeepSeek V4 Flash (model id deepseek-v4-flash) is $0.2778 / $1.04, and DeepSeek V4 Pro (model id deepseek-v4-pro) is $1.11 / $3.33. GLM-5.3 (model id glm-5.3) is $0.5556 / $1.67. Kimi K3 (model id kimi-k3) is $1.04 / $5.21. Qwen3.7 Plus (model id qwen3.7-plus) is $0.1278 / $0.4167, and Qwen3.8 Max (model id qwen3.8-max) is $0.6944 / $2.08. MiniMax M3 (model id MiniMax-M3) is $0.1389 / $0.5556. MiMo V2.6 Pro (model id MiMo-V2.6-Pro) is $0.2083 / $0.3472.

These OpenAI-format models can be called from Cursor, Cline, Roo Code, OpenCode, Cherry Studio, any OpenAI SDK or cURL, all against https://api.vanaapi.com. Keeping several families on one address means a project can move between them without a new account, a new key or a new configuration file.

§ Balance, keys, usage log and the small print

Vana API bills from a prepaid balance, charged per real token, with no monthly fee on the pay-as-you-go side. Keys are created in the dashboard on https://vanaapi.com, and every call is listed in the usage log there. That combination is what makes a heavy week manageable: the balance tells you what is left, and the log tells you where it went.

Signing up adds $0.10 of credit to the account. The referral programme pays 14% of what people you invite pay, and 5% one level further; earnings can be withdrawn from $13.89. If a top-up is not what you expected, unused balance can be refunded within 3 days of the top-up, minus what was used and the payment processor's fee.

For a first month, the sequence is short. Create a key, top up, send a test call to https://api.vanaapi.com with the model id you need, and watch the first entry appear in the usage log. If the week ahead is GPT-heavy, the GPT plan at $103.61 for 30 days with a $1,200 weekly cap gives you a defined ceiling. If the week ahead is Claude-heavy, the $64.72 plan with a $320 weekly cap or the $153.61 plan with a $750 weekly cap gives you the same kind of limit. One address, three protocols, minimal setup: that is the shape of Vana API.