Curated Model Catalogue
Five to seven proven open-weight models in v1. Foundation models (Llama, DeepSeek, Qwen). Clinical specialists (MedGemma, pathology vision transformers). An embedding model for retrieval. A speech model for transcription. Each one validated end-to-end on Bifrost infrastructure, version-pinned per playbook deployment, refreshed quarterly.
The catalogue is small on purpose. Curation is the moat.