One LLM API. 30+ open-source models. Pay per token.
Call Llama, DeepSeek, Qwen, Kimi, GLM, Mistral and more through a single OpenAI-compatible API, served from our sovereign AI cloud: our own GPUs in Paris and Lyon. Nothing to deploy, no GPU to manage.
OpenAI-compatible · Hosted in France · Free playground: 10 free requests
import os
from openai import OpenAI
client = OpenAI(
base_url="https://console.arkanecloud.com/api/v1",
api_key=os.environ["ARKANE_CLOUD_API_KEY"],
)
completion = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4.1-Flash",
messages=[{"role": "user", "content": "Hello!"}],
)
Get started with the Models API in three steps.
01
Test in the playground
Compare models on your own prompts in the console playground, before writing a single line of code.
02
Create your API key
Top up your balance and create a key in the console. You only pay for the tokens you actually use.
03
Change the base URL
Point your OpenAI SDK to Arkane Cloud. Same code, same request format, models hosted in France. See how it fits your AI inference workloads.
The best open-source models, ready to call.
Chat, reasoning, coding and image generation models, updated as new open-source releases land. Pick a model ID and send a request.
Everything you need to ship AI features.
A single endpoint for the best open-source models, with the compatibility, cost control and data residency your team expects, from chat features to AI agents.
Compatibility
OpenAI-compatible API
Use the official OpenAI Python and JavaScript SDKs. Migrating takes a base URL and API key change, not a rewrite.
Simplicity
Nothing to deploy
Models are already running behind one endpoint. Choose a model ID and send a request: no GPU to size, no endpoint to configure.
Pricing
Transparent per-token billing
Input, output and cached tokens are billed per model and deducted from your prepaid balance. No subscription, no minimum.
Visibility
Usage and cost tracking
Follow tokens and costs in the console, filtered by model, API key or period, so you always know where your budget goes.
Sovereignty
Your data stays in France
Requests are processed on our own NVIDIA GPU servers in Paris and Lyon. Your prompts and outputs are never used to train models.
Performance
Optimized inference stack
FP8/FP4 quantization, tuned vLLM and proprietary CUDA kernels deliver low latency and high throughput on every model.
Coding with AI? Use the Models API in VS Code chat and agent mode with the free Arkane Cloud extension → · See the Coding use case
Ready-made models or your own model?
Both give you an OpenAI-compatible API hosted in Paris and Lyon. The difference is what you pay for and how much control you get.
Models API
Serverless GPU
You get
A shared API to our model catalog
Your own endpoint, running the model you choose
Models
30+ ready-to-call open-source models
Any catalog model, or any vLLM-compatible Hugging Face model, including fine-tuned ones
You pay for
Tokens (input, output, cached) or images
GPU time, billed per second
GPU choice
Managed by Arkane Cloud
You choose the GPU
When idle
$0, you only pay per request (by tokens)
$0 with scale-to-zero
Best for
Prototyping, variable traffic, testing many models
Custom or fine-tuned models, long prompts, steady traffic
Models API pricing: pay per token, nothing else.
Each model has its own per-million-token rates. You only pay for the tokens you use, deducted from your prepaid balance, and you can follow costs in real time in the console.
See model prices →Your prompt, system instructions and conversation history.
The tokens generated by the model.
Input read from the prompt cache, billed at a lower rate when the model offers one.
Image generation models are billed per image.
Questions about the Models API
Is the Models API compatible with OpenAI?
Yes. Use the official OpenAI Python or JavaScript SDK with the base URL https://console.arkanecloud.com/api/v1 and your Arkane Cloud API key. In most codebases, nothing else changes.
Which API features are supported?
Streaming via server-sent events (SSE), plus tool calling and JSON / structured output through the standard OpenAI-compatible parameters. Tool calling and structured output support depends on the model. Vision models also accept image input.
Where is my data processed?
On our own NVIDIA GPU servers, hosted in colocation datacenters in Paris and Lyon, France.
Are my prompts used to train models?
No. Your prompts and outputs are not used to train any model, neither ours nor anyone else's.
How does billing work?
You pay per million tokens, with separate rates for input, output and cached input depending on the model. Image models are billed per image. Usage is deducted from your prepaid balance, with no subscription or minimum.
Can I try it for free?
Yes. The console playground lets you test models for free, with 10 free requests. The free tier applies to the playground only, not to API calls.
Can I get a free API key?
The console playground includes 10 free requests, so you can compare models before writing code. API usage is billed per token and deducted from your prepaid balance: add funds in the console before your first API request.
Can I run my own or a fine-tuned model?
The Models API serves our catalog. To deploy any Hugging Face model, including fine-tuned ones, use Serverless GPU and pay by the second.
Are there request limits?
A request body can be up to 32 MiB, including text and tool definitions. Context length depends on the model: check each model in the model catalog.
Send your first request today.
Create your account, test models in the playground, then switch your base URL.