Custom models
A custom model is a versioned workspace resource that combines a base model with optional saved questions and task-specific improvements, so you can run and refine predictions through a single model name.
This guide creates a custom model that flags urgent support tickets. Replace acme with your workspace and set up authentication before you start.
Supported base models
Use mo-latest for the current base model or mo-2.0 to pin its version.
| Model | Alias | Context window | Modalities |
|---|---|---|---|
mo-2.0 | mo-latest | 256K tokens | Text, images |
mo-1.3.1 | — | 256K tokens | Text, images |
mo-lite Coming soon | — | — | — |
Mo Lite is coming soon. Its API requests currently return 400 model_not_supported.
See List models for the models and repositories available to your token.
1. Save the question
Create the model with a base model and a named question. Keep the question ID stable so future answers and annotations refer to the same task.
from modaic import Modaic
client = Modaic()
model = client.models.create(
workspace="acme",
slug="support-priority",
model="mo-latest",
questions={"urgent": {"type": "noul", "instructions": "Does this require immediate attention?"}},
)
2. Run the custom model
Send a new ticket to acme/support-priority. The model loads its saved question and returns the answer under urgent.
result = client.predictions.create(
model="acme/support-priority",
state="Checkout is unavailable.",
)
print(result.answers["urgent"].noul)
If the response includes an example_id (exampleId in TypeScript), the input was captured for review. Use reviewed examples to optimize the question's instructions.
See Versioning to pin a revision, or List models to find available base models.