Skip to content
MAKINA

Platform/Models

Hosted Models, or Your Own

In Makina's cloud, choose a model from the catalog for each task, or leave it on the default. In a private deployment, Operator runs your own open-weight models on your own GPUs.

Request access

The catalog, as it is today. More models are added as they qualify.

  • GPT-5.6 Luna

    Default · Served by OpenAI

    The default: Operator uses it unless you choose another. Made by OpenAI and served by OpenAI's own API.

  • Qwen3.8 Flash

    Served through OpenRouter

    Made by Qwen, Alibaba's family of models, and served through OpenRouter.

  • Kimi K3

    Served through OpenRouter

    Made by Moonshot AI and served through OpenRouter.

  • Grok 4.7

    Served through OpenRouter

    Made by xAI and served through OpenRouter.

  • GPT-5.6 Sol

    Served through OpenRouter

    Made by OpenAI and served through OpenRouter.

Choose a model in the task box, or leave it on the default. The menu shows a catalog model’s price before you choose it, and the Pricing page lists them all.

Private deployments

Your Own Models, on Your Own GPUs

In a private deployment, Operator runs open-weight models on GPUs you own. We help you choose them, and test them on your own tasks before they do real work.
Private deployments

Questions About Models

Which model does the work, and how to choose another.

Model evaluation
Which model does Operator use?
GPT-5.6 Luna, on OpenAI, unless you choose another model for a task.
How do I choose another?
Open the model menu in the task box and pick one from the catalog. The choice holds for that task.
How are models added?
The catalog grows as models qualify for the work Operator does. New ones appear in the model menu.
Can we run our own model?
Yes, in a private deployment: open-weight models on your own GPUs. Talk to us.
Does a model learn from my work?
We do not train models on your work, and we ask OpenAI not to store what it is sent.