Pay per token, or license the platform.
ARK Cloud is prepaid and metered: you pay for the tokens you use, nothing else. ARK Core and ARK Tailored are licensed per GPU for your own infrastructure.
Every model, every price.
US dollars per million tokens, or per audio hour for speech. The same price applies whether a request is streamed or not.
| Model | Type | Input | Output |
|---|---|---|---|
| Qwen3.6-35B-A3Bper 1M tokens | Chat Vision, Reasoning, Tools | $0.0156 | $0.4499 |
| Qwen3.5-122B-A10Bper 1M tokens | Chat Vision, Reasoning, Tools | $0.038 | $1.3902 |
| Qwen3.8-Flash-Nextper 1M tokens | Chat Vision, Reasoning, Tools | $0.057 | $2.0853 |
| Qwen3.8-27Bper 1M tokens | Chat Vision, Reasoning, Tools | $0.038 | $1.3902 |
| Qwen3.5-4Bper 1M tokens | Chat Vision, Tools | $0.0097 | $0.3268 |
| Bielik-11B-v3.0-Instructper 1M tokens | Chat Polish-first | $0.10 | $0.40 |
| bge-m3per 1M tokens | Embeddings | $0.0002 | — |
| whisper-large-v3-turboper audio hour | Speech to text | $0.0018 | — |
Image input is billed as prompt tokens at the input price, each image token counted 10 times: an image of 1,024 image tokens is billed as 10,240 input tokens. A typical photo is 256 to 1,280 image tokens.
Reasoning tokens count as output. Cached prompt tokens make responses faster and are billed at the normal input price.
Each request is rounded up to a whole credit. Requests that fail before producing output, model listings and token counting are free.
Prepaid, with no subscription.
One million credits cost one dollar; a price of $0.45 per million tokens is 0.45 credits per token. Buy a pack on the ARK Cloud portal and every model is available at the same prices.
New accounts receive free starter credits on their first sign-in with Google. Payment is by card through Stripe; VAT is added at checkout where it applies. Every pack gives the same access: there are no tiers or rate limits by plan.
A few worked examples.
Computed from the list prices above.
| Workload | Model | Cost |
|---|---|---|
| 1,000 summaries of 1,100-token documents, 100-token answers | Qwen3.6-35B-A3B | $0.062 |
| The same 1,000 summaries | Qwen3.5-122B-A10B | $0.181 |
| Embedding 10 million tokens of documents | bge-m3 | $0.10 |
| Transcribing 100 hours of recordings | whisper-large-v3-turbo | $1.44 |
| A 50,000-token contract read and answered in 2,000 tokens | Qwen3.5-122B-A10B | $0.0047 |
Licensed per GPU, on your infrastructure.
Self-hosted ARK is priced by the GPUs it runs on. You pay no per-token fees: run as much traffic as your hardware carries. Quotes are prepared for your fleet and models.
Per GPU
Every GPU in the deployment, from an RTX card to a data-centre accelerator, counts as one licence. Includes up to 10 models and text generation.
+€10 per GPU per month
For each added modality: vision, embeddings or speech to text.
+€25 per GPU per month
For deployments that serve 11 to 20 models.
Support by fleet size
| Level | GPU licences | Channel | Hours |
|---|---|---|---|
| Standard | fewer than 25 | Business hours | |
| Premium | 25 to 74 | Email and chat | Extended hours |
| Enterprise | 75 or more | Dedicated technical account manager | 24/7 for critical issues |
ARK Core
You deploy and operate ARK yourself with our documentation and release packages, and bring your own identity, logging and monitoring.
About ARK CoreARK Tailored
We plan and deploy ARK with your team: capacity, models, identity and observability, then hand over a platform your engineers run.
About ARK TailoredPricing FAQ
Why are input tokens so much cheaper than output tokens?
A model reads your whole prompt in one parallel pass, but writes its answer one token at a time. Generating a token takes far more GPU time than reading one, and the prices follow that.
What happens when my credits run out?
Requests return 402 until you top up on the portal. Nothing is charged automatically.
Can I see what each request cost?
Yes. Every response's usage includes credits_used and your credit_balance. See Usage and billing.
Can I get an invoice?
Stripe sends a receipt for every purchase. For invoices, write to support@ark-labs.pl.
Can I get reserved capacity or a dedicated deployment?
Yes. With ARK Core you run a dedicated deployment on your own GPUs; with ARK Tailored we size and deploy it with you. Talk to us.
Start with free starter credits.
Sign in with Google and make your first request in minutes.