Frequently asked questions.
Everything about Arkane Cloud, the Sovereign AI Cloud: products, billing, GPUs and data. Can't find your answer? Our team is here to help.
Getting started
What is Arkane Cloud?
Arkane Cloud is a Sovereign AI Cloud. We are a French company that runs its own NVIDIA GPU servers in colocation data centers in Paris and Lyon. We offer four products: the Models API, Serverless GPU, Cloud GPU and GPU Clusters. Learn more about us.
How do I get started?
Create an account in the console, test models in the free playground (10 free requests), then create an API key and call the Models API from your code. Step-by-step guides are in the documentation.
Which product should I choose?
Choose the Models API to test and ship fast, billed per token. Choose Serverless GPU to run your own or a fine-tuned model, billed per second with scale-to-zero. Choose Cloud GPU for full control of GPU instances, from 1 to 8 GPUs with SSH access. Choose GPU Clusters for dedicated capacity reserved for 3 months to 5 years. Not sure? Take the free AI maturity assessment.
Is the API OpenAI-compatible?
Yes. Set the base URL to https://console.arkanecloud.com/api/v1 and use your API key in any OpenAI SDK. Streaming is supported; tool calling and JSON / structured outputs depend on the model. See our pages on agentic AI and AI for coding.
Can I use Arkane Cloud in VS Code?
Yes. Install the free Arkane Cloud extension for chat and agent mode, or use any tool that accepts an OpenAI-compatible base URL. See AI for coding.
Products & GPUs
Which models are available?
More than 30 open-source text and image models through one API, including the DeepSeek, Qwen, Kimi, GLM, Llama, MiniMax and GPT-OSS families. See the full list and prices on the pricing page.
Which GPUs do you offer?
Serverless GPU and GPU Clusters run on our own NVIDIA B300 (288 GB HBM3e) and RTX PRO 6000 Blackwell (96 GB) servers in Paris and Lyon. Cloud GPU instances, hosted in Europe, offer a wider range of NVIDIA GPUs.
Can I deploy my own or a fine-tuned model?
Yes. On Serverless GPU, you can deploy a catalog model, any vLLM-compatible Hugging Face model (public or private) or your own container, behind an OpenAI-compatible endpoint. See fine-tuning.
How does autoscaling work on Serverless GPU?
You set a minimum and a maximum number of replicas. Replicas are added as requests queue up and removed when demand falls. Set the minimum to 0 to scale to zero; the first request after an idle period then waits for the model to start.
What are GPU Clusters?
Dedicated multi-node NVIDIA clusters with InfiniBand, deployed with managed Kubernetes or on Bare Metal. Clusters are single tenant and reserved for 3 months to 5 years, depending on the GPU. Request a quote.
What storage is available?
On Cloud GPU, you choose an OS image and attach block volumes, which you can detach and reuse across instances. Storage is billed per GB per month (see pricing).
Billing
How is usage billed?
Each product has its own billing model: the Models API is billed per million input, output and cached tokens (images per image), Serverless GPU per second while replicas run, Cloud GPU per second from start to stop, and GPU Clusters on a reserved commitment. Only successful API requests are billed. See pricing.
Is there a free trial?
Yes. The playground includes 10 free requests to test models, with no credit card required. Create your account to get started.
Are there setup fees or minimum commitments?
No setup fees and no minimum spend on pay-as-you-go products. Reserved capacity and GPU Clusters come with a commitment of 3 months to 5 years in exchange for better pricing. The minimum term depends on the GPU and the product.
Do you offer volume or reserved pricing?
Yes. We offer reserved capacity and custom pricing for high volumes. Contact sales to discuss your needs.
How do I pay?
You add credit to your account in the console, and usage is deducted from your balance. Prices are in USD.
Security & data
Where is my data processed?
The Models API, Serverless GPU and GPU Clusters run on our own servers in Paris and Lyon, France. Cloud GPU instances are hosted in Europe. Arkane Cloud is operated by a French company.
Are my prompts and data used to train models?
No. Your prompts and outputs are never used to train models.
Are GPU Clusters dedicated to me?
Yes. GPU Clusters are single tenant: the nodes are reserved for you, with managed Kubernetes or Bare Metal deployment.
How can I talk to your team?
Contact sales for a project, a budget or reserved capacity, or request a cluster quote.
Still have questions?
Our team can review your project, your usage and your budget with you.
Free playground: 10 free requests · No commitment