The ARK API, hosted in the EU.
Serverless inference on open-weight models, behind the OpenAI and Anthropic APIs your code already speaks. Sign in with Google, take your key, pay per token from prepaid credits.
Free starter credits on first sign-in. No card needed to start.
curl https://api.ark-labs.cloud/api/v1/chat/completions \
-H "Authorization: Bearer $ARK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "arklabsai/Qwen3.5-122B-A10B",
"reasoning_effort": "high",
"messages": [
{"role": "user", "content": "Plan a DORA readiness review in five steps."}
]
}'response = client.chat.completions.create(
model="arklabsai/Qwen3.5-122B-A10B",
reasoning_effort="high",
messages=[{"role": "user", "content": "Plan a DORA readiness review in five steps."}],
)
print(response.choices[0].message.content)- Models online
- 6chat, vision, reasoning, embeddings, speech
- Context
- 256K tokenson the largest Qwen models
- From
- $0.0097per million input tokens (Qwen3.5-4B)
- Hosted
- In the EUon GPU servers we operate
A complete inference API, without the infrastructure.
OpenAI-compatible
Chat completions, embeddings, audio transcriptions and model listing at /api/v1. The OpenAI SDKs work as they are.
Anthropic-compatible
Messages and token counting at /api/anthropic/v1, with thinking, tools, images and documents.
Reasoning and tools
Turn on thinking per request; give models your functions with JSON Schema, streamed or not.
Vision
Send documents, screenshots and photos to the Qwen models together with your question.
Stateful sessions
Keep a conversation on the server and send only the new turn. You pay only for new tokens.
Cost in every response
Each response reports its tokens, the credits it used and your remaining balance.
Four steps to your first token.
ARK Cloud is self-service. No sales call, no contract, no minimum spend.
Sign in with Google
Open the ARK Cloud portal and continue with your Google account. Your account comes with free starter credits.
Copy your API key
The key is on your dashboard, ready to use with both APIs.
Point your SDK at ARK
Set the base URL to https://api.ark-labs.cloud/api/v1 and choose a model.
Top up when you need to
Buy prepaid credit packs of $5, $50 or $500 by card. Every pack unlocks every model.
European infrastructure, by default.
Inference on ARK Cloud runs on GPU servers we operate in the European Union. We serve models; we do not train them, and we do not train anything on your data.
- Logs without content. Request logs keep metadata such as time, model and credits used. Prompt fields are redacted and generated text is not logged.
- Open-weight models only. You know exactly which model answered, and under which licence.
- A path in-house. The same API runs on your infrastructure with ARK Core or ARK Tailored, so moving workloads home is a configuration change in your code.
ARK Cloud FAQ
Is there a free tier?
New accounts receive free starter credits on their first sign-in. After that, ARK Cloud is pay-as-you-go from prepaid credits.
Are there rate limits?
There are no per-key request quotas. When a model is at capacity, new requests are refused with 429 and should be retried with backoff. See Limits and capacity.
Can I use my existing OpenAI or Anthropic code?
Yes: change the base URL, the API key and the model id. ARK serves chat completions, embeddings, transcriptions and Anthropic messages; endpoints such as Assistants, Batches or Files are not available.
Can I chat with the models without writing code?
Yes. ARK Chat is a browser chat on the same models, signed in with the same account.
Can I run my own fine-tuned model?
Not on ARK Cloud today. With ARK Core you run the models you need on your own infrastructure: fine-tunes of the model families ARK supports run like the originals.
Your first request is two minutes away.
Sign in with Google, copy your key, call a model.