EU-hosted · Zero retention · No credit card

Pay per token. Nothing else.

Transparent per-token rates on 10+ frontier open models. Credits never expire. No auto-renewal, no hidden fees, no credit card required to start. Hosted in the EU with zero retention by default.

Start free → Talk to sales

Free credits included — sign up, connect your SDK, start building in minutes.

$1 = 1M
Credit
Exchange
∞
Credits
Never Expire
100%
EU Data
Residency
0
Retention
By Default
USD IN $1 becomes CREDIT POOL 1,000,000 credits · never expire spend anywhere 💬 TEXT LLMs, reasoning 🖼 IMAGE SD 3.5 🎤 SPEECH Whisper 🧲 EMBED RAG, search THE EXCHANGE RATE $1 = 1,000,000 credits prices in USD · same math at any scale
PREPAID PACKAGES · ONE CURRENCY
Load the credits you need.
Use them whenever.

ARK Cloud runs on a single pool of credits you can spend across every model and every modality — text generation, embeddings, image generation, speech recognition. Load once, spend anywhere. Credits never expire. Nothing auto-renews. Three prepaid packages below, or pay as you go after the free credits.

5M credits
Never expire · spend any modality
Jumpstart
Explore
$5 USD
  • First-look at 10+ frontier open models
  • OpenAI v1 / Anthropic-compatible API
  • Portal UI for chat, images, embeddings
  • EU-hosted, zero retention by default
  • Best-effort 99% availability
Sign up →
50M credits
Never expire · spend any modality
Momentum
Build
$50 USD
  • Everything in Jumpstart
  • Workflow automation & MVP workloads
  • Higher concurrent throughput on shared endpoints
  • Image generation & speech recognition
  • Credits usable across every model & modality
Sign up →
500M credits
Never expire · spend any modality
Command
Ship
$500 USD
  • Everything in Momentum
  • Production-grade inference workloads
  • Higher sustained throughput on shared endpoints
  • Priority onboarding & integration support
  • Path to dedicated endpoints on request
Sign up →
PER-MODEL RATES
Every model. Every rate. Public.

No enterprise-only pricing schedules. The rate you see is the rate you pay. All prices are per 1 million tokens unless stated otherwise.

Large Language Models · stateless

USD per 1M tokens · input / output billed separately
ModelInput (per 1M tokens)Output (per 1M tokens)
meta-llama/Llama-3.1-8B-InstructGeneral-purpose instruction model $0.01per 1M input tokens $0.49per 1M output tokens
Qwen/Qwen3-32BAdvanced reasoning & instruction following $0.01per 1M input tokens $3.99per 1M output tokens
Qwen/Qwen3-Coder-30BCode generation & software reasoning $0.01per 1M input tokens $3.99per 1M output tokens
speakleash/Bielik-11B-v3.0Polish-first language model $0.01per 1M input tokens $0.99per 1M output tokens
openai/gpt-oss-20bOpen-weight GPT family $0.01per 1M input tokens $0.99per 1M output tokens

Embeddings, image, and speech

Flat per-unit pricing — credits from the same pool
ServiceModelRate
Text embeddingsDocument retrieval, RAG, semantic search BAAI/bge-m3Multilingual $0.01per 1M tokens
Image generationText-to-image & image-to-image Stable Diffusion 3.5Frontier open image model $0.019per image
Speech recognitionMultilingual transcription WhisperIndustry-standard ASR $0.0144per hour transcribed

Custom models · on request

Open-source fine-tunes & proprietary models
Hosted custom modelsYour weights on our infrastructure — scoped per engagement Contact sales →
WHAT'S INCLUDED EVERY TIME
The commitments behind every credit.
🇪🇺

EU-only data residency

Every request is served from infrastructure inside the European Union. Data never transits, never stores, never lands outside the region.

🔒

Zero retention by default

Prompts and completions are not stored. Not used for training. Not reviewed by humans. Enable logging only if you explicitly opt in.

✅

EU AI Act ready

Transparency documentation, audit logs, and data processing agreements available for enterprise customers. Ready for regulated use from day one.

⏳

Credits never expire

No 12-month burn window. No rollover gotchas. Load credits when you need them. Spend them when the project actually lands.

Free credits. No credit card.
Start building in minutes.

Sign up, get credits, point your OpenAI-compatible SDK at ARK Cloud. The first request takes about 90 seconds from cold start.

Start free → Read the docs
Running inference at volume, or on your own hardware?
At a certain workload, per-token billing flips and a per-GPU license is the cheaper path. That's what ARK Tailored and ARK Core are for — the same runtime, deployed on your infrastructure.
See Enterprise pricing →