We finetune an open-weight model on your data, privately - and you own the result. The more of the work you hand us, the higher the tier. Never a per-token meter. Self-serve and a public library are coming soon.
Every tier is private: your data stays yours, and we keep none of it.
Train free - you share it back.
$0 / forever
Your data, your GPU. We take nothing.
$20 / model
No GPU? Send the data, we train it - privately.
$99 / model
You have the data but not the plan? We dig in.
$999 / project
When your data flows from ongoing work, we train continuously: weekend batches that fold in the week's data, scaling up to a dedicated engineer and a GPU cluster on your own hardware. From steady operations to a fully dedicated setup - tell us the shape and we'll scope it.
The real alternative to letting us build it isn't another subscription - it's hiring someone who can. Here's the gap.
Specialist salaries vary by region (~€65-85k/yr in Eastern Europe, ~€150-230k/yr in the West, fully loaded). For a small business the choice usually isn't "cheaper" - it's possible vs. impossible. And we never meter your questions: the flat price is the whole bill.
Run your own finetuned models locally for everyday work. Escalate to a frontier model only for the ~1% of genuinely hard cases - and pay for just those.
Tell us what you want to build