AIOpenAI
Temporarily unavailable
Model ID
openai/gpt-5
GPT-5
Every AI model.
One key.
Pay in your currency
GPT, Claude, Gemini, DeepSeek and Qwen through a single OpenAI-compatible API. One balance, one invoice, no juggling five vendor accounts.
$1.00 of free credit when you sign up. No card needed.
Available
with one key
- GPT-5
- Claude Sonnet 4.5
- Gemini 2.5 Pro
- DeepSeek V3.1
- Qwen3 Coder
- GPT-5 mini
- Claude Haiku 4.5
- Gemini 2.5 Flash
- gpt-oss-120b
- Kimi K2.6
- Llama 3.3 70B
How it works
Three steps to your first response
sk-harvanera-…Create key
Step 1
Create a key. Email or Google — and you're in.
base_url = "harvanera.com/v1"
AIHi! How can I help? $0.0004
Step 2
Swap the base_url. OpenAI SDK, Cursor, n8n — no code changes.
Step 3
Watch your spend. By key, by model, by day.
Moved off five vendor accounts in one evening
“We changed the base_url and everything just worked. The real win is billing: finance gets one invoice a month instead of chasing five.”
— Anna Krylova, CTO, chatbot studio
Works with your tools
Your key works anywhere you can set an OpenAI-compatible base_url: IDEs, agents, no-code tools and chat clients.
- Cursor
Cline
LangChain
n8n
Ollama
JetBrains AI
Zed
Obsidian
Models we keep in shape
AnthropicTemporarily unavailable
Model ID
anthropic/claude-sonnet-4.5
Claude Sonnet 4.5
GoogleTemporarily unavailable
Model ID
google/gemini-2.5-pro
Gemini 2.5 Pro
DeepSeekModel ID
deepseek/deepseek-v3.1
DeepSeek V3.1
AlibabaModel ID
qwen/qwen3-coder
Qwen3 Coder
AIOpenAI
Temporarily unavailable
Model ID
openai/gpt-5-mini
GPT-5 mini
AnthropicTemporarily unavailable
Model ID
anthropic/claude-haiku-4.5
Claude Haiku 4.5
GoogleTemporarily unavailable
Model ID
google/gemini-2.5-flash
Gemini 2.5 Flash
AIOpenAI
Model ID
openai/gpt-oss-120b
gpt-oss-120b
AIMoonshot AI
Model ID
moonshotai/kimi-k2.6
Kimi K2.6
AIMeta
Model ID
meta-llama/llama-3.3-70b
Llama 3.3 70B
Flagship · OpenAI
The strongest all-rounder: code, analysis, agents
≈ $0.0063 per answer · 82 tok/s · context 400K
Features
What Harvanera takes off your plate
So you can think about your product, not about model access.
prod-backend
$200
intern-cursor
$15
old-bot
—
Limits on every key
Budget, model allow-list and IP rules. An intern's key can't burn the monthly budget overnight.
your request
source A · 429
source B · 200
Fallback sources
Every model has at least two upstreams. If one fails, the request goes to another — you just see 200 OK.
Invoice
Statement
Receipt
Built for teams and companies
Invoices, monthly statements and one shared balance for the whole team.
For vibe coding
Cursor, Cline and Roo Code in two minutes.
Ready-made configs for IDEs and agents: paste one line and the model is writing code through your key.
Base URL: https://harvanera.com/v1 API key: sk-harvanera-… Model: qwen/qwen3-coderOpen the guides
Detailed analytics
Spend by day, model and key, with CSV export. See what works and what could run on a cheaper model.
Pricing
Pay only for tokens
No subscriptions. Top up your balance in your currency, and each request is charged at the model's fixed price. Unused balance can be refunded on request.
- Price per 1M tokens and per typical answer
- A receipt for every top-up
- An email when your balance runs low
FAQ
Questions people actually ask
So what am I actually paying for?+
For tokens — the text a model reads and writes. You top up once, and every request is deducted at the price listed next to the model. No subscription, no minimum monthly spend, no surprise overage.
Will I have to rewrite my code?+
No. If your app talks to the OpenAI API, change two lines — base_url and the key — and you're done. The same goes for Cursor, Cline, n8n, LangChain and anything else that accepts a custom endpoint.
Which countries and payment methods do you support?+
We work with customers worldwide. You can pay by card in eight currencies, and companies can ask for an invoice. Your balance is kept in USD, so the price of a request doesn't jump around with exchange rates.
What if a model goes down on the provider's side?+
Most models have a backup source, so the request is quietly retried there. If there's no healthy source left, you get a clear error and we don't charge you for it.
Do you read or keep my prompts?+
No. By default we store only metadata: which model, how many tokens, what it cost. Your prompts go to the model provider to generate the answer, and that's it. You can turn on full logging for debugging if you want it.
Can I get my money back?+
Yes. Unused balance is refunded on request. If you're in the EU or UK, you also have a 14-day right of withdrawal — the details are in our refund policy.
How do I try it before paying?+
Sign up and you get free credit straight away — enough for a few hundred answers from a budget model. No card needed.
Why not just go to OpenAI or Anthropic directly?+
You can. We're useful when you want several vendors without several accounts, cards and invoices; when you need a budget per key; or when you want automatic fallback if one provider is having a bad day.