Lower Inference Costs
Through our direct relationship with a world-class inference provider, we pass on significant savings on the compute that powers your AI — running the same SOTA frontier and open models, without compromising on model quality or performance.
Flat, Fixed Pricing
Your AI runs on a flat, fixed rate. Powered by leading frontier, open-source models, including DeepSeek V4-Flash, Qwen 3.6, GLM 5.3-Flash.
No Overages, No Lock-In
No usage traps, no overage fees, no bill shock. What you’re quoted is what you pay — and you can walk away whenever you choose.
Local, open-weight models — deployed where your data lives
Every plan runs on sovereign inference. We run frontier open-weight models on hardware you own, or on dedicated zero-log infrastructure segregated for you: Qwen, DeepSeek, Llama, Mistral and GLM. No prompt, file or embedding leaves your perimeter, and nothing is billed per token.
A private inference layer: model selection and benchmarking on your own tasks; deployment on-premises, in a private cloud or at the edge; retrieval (RAG) over your corpus; inference operations, monitoring and evaluation; and an OpenAI-compatible API your existing tools connect to unchanged. Air-gap capable, zero-log by policy with access control, and compliant by construction. Flat, not metered, with no vendor leverage: the models are yours to swap or fine-tune, so there is no lock-in.
Market signal: enterprise inference running on-premises or at the edge has grown from ~12% in 2023 to ~55% today, and on-premises deployments run at a fraction of premium cloud API cost per token. Sovereignty and cost are the two reasons buyers give.
Compliance is not what this price trades away. GDPR, HIPAA, SOC 2, ISO 27001/42001 and the EU AI Act are met by construction. How we secure it
The same 5 billion tokens, two ways.
A frontier model meters every token you send. We charge the month. One month of the same workload, priced both ways.
Each rule is that bill’s share of the largest one on the table.
Monthly cost of 5 billion input tokens, in US dollars, against our Starter plan at 399 euros a month. Rows run from the highest list price to the lowest.
| Model or plan |
Per million tokens |
At 5B tokens a month |
Against our flat plan |
| ChatGPT · GPT-6 Astra |
$10.00 |
$50,000 |
+$49,553
112× our plan |
| Claude · Opus 5.5 |
$4.00 |
$20,000 |
+$19,553
45× our plan |
| Google · Gemini 3.1 Pro |
$2.00 |
$10,000 |
+$9,553
22× our plan |
| DeepSeek V4 Flash, open weights |
$0.14 |
$700 |
+$253
1.6× our plan |
| Our flat planStarter, the five-billion-token tier |
not metered |
$447 |
baseline
1× our plan |
The table counts input tokens only, so every figure in it is a floor, not a ceiling. List prices are dated 9 October 2026 and are re-verified before every print: promotional rates expire and models leave price pages. The strongest ground here is the premium closed models at the top of the table — the models most businesses are already paying for.