HYBT Cloud

A private cloud for open models.

Pick an open-weight model, get an OpenAI-compatible API, fine-tune on your own data, and set a budget that stops spending when it should. HYBT Cloud is in early access.

Request early access See what is included
  1. modelopen-weight, 27B class
  2. gpu tierWorking · 96 GB
  3. apiOpenAI-compatible
  4. budgetdaily limit, stops at the limit
  5. idlereleased when unused
  6. datastays in your workspace
Example of a deployment. It is not a price list.

What is included

  • Model endpoints

    Open-weight models behind an OpenAI-compatible API. The catalog grows during early access, and every model's licence is checked for commercial use.

  • Fine-tuning workspace

    JupyterLab with LoRA and QLoRA templates, your dataset, an evaluation set, and adapter export.

  • Document assistant

    A ready template that answers questions over your own documents, with an evaluation set to check the answers.

  • Keys and budgets

    API keys per team, spending limits, automatic stop, and a usage log.

  • Sovereign Bridge, in pilot

    For regulated clients: policy-controlled access to approved cloud services, minimised payloads, and an audit trail.

  • Help choosing

    Advice on the model and the GPU tier for your workload, from the team that runs these workloads.

GPU tiers

Small
RTX 4000 Ada, A6000 and 4090 class. Models up to about 13B parameters, light fine-tuning, and development.
Working
RTX PRO 6000 with 96 GB of memory. The main tier for large open models in 4-bit and for fine-tuning.
Large
A100 and H100. Heavier fine-tuning and high-throughput serving.
On request
H200 and B200.

Capacity is provided through infrastructure partners. Availability and pricing are confirmed per request during early access.

Request access

Fill in the form and your email app opens with the message ready to send. Nothing is stored by this site. We reply within two business days.