Skip to main content
The Models page separates Axionic-hosted foundation models from your own trained models.

​
Hosted models

Hosted models are the shared foundation models Spectra can use for training targets and runtime workflows. They are managed by Axionic and appear separately from anything you train yourself. Use hosted models when you want to:
  • choose a student target for fine-tuning
  • pick a runtime model inside the Optimization workspace
  • work with a model that already supports the features you need, such as SAE-backed behavior monitoring

​
Your models

Your model library includes:
  • active training runs
  • ready text and tool models
  • completed speech recognition and OCR training artifacts
  • saved drafts that have not been launched yet
Trained models are stored privately. Your current private-model allowance depends on your account tier and is shown on Billing. Custom model icons are stored with the training run and follow the model into the Models page and the Optimization workspace. If a trained model is served as a hosted endpoint, Spectra still uses the training-run icon metadata when it renders that hosted model for the owning user.

​
Common actions

  • Open details to review identifiers, training metadata, and integration values when a model supports API use
  • Resume from draft to continue a saved training configuration
  • Cancel an in-progress run
  • Delete a failed run or unused draft
  • Use in optimization for ready text and tool models that support runtime steering, testing, and sampling controls

​
What this page does not include

This page is for inventory and model lifecycle management. Runtime steering, vector controls, testing, and API snippets for API-ready text and tool models live under Optimization. For the training flow itself, see Training. For integration examples after a text or tool model is ready, see Using Your Model.