Passa al contenuto principale

Types of models

In Rational AI you can add three types of models: Fine-tuned, Open source and Third party.

1) Fine-tuned model​

A model that’s been trained further on your own data (or a specific dataset) to behave a certain way—tone, domain knowledge, formats, policies, etc.

  • Best for: consistent behavior for a specific use case (support, internal tooling, company style).
  • What you usually manage: which fine-tune to use, and its base model lineage.

2) Open source model​

A model you download and run yourself (commonly from Hugging Face), often as a local file format like GGUF with a chosen quantization.

  • Best for: running on your own infrastructure, cost control, offline/air-gapped use, customization.
  • Tradeoffs: you own the hosting + compute + storage; model sizes can be large.

3) Third-party model (proxy)​

A model that’s hosted by an external provider (e.g., OpenAI / Anthropic / Google). Your platform stores a configuration that routes requests to that provider.

  • Best for: fast setup, strong quality, no hosting burden.
  • Key concept: you don’t “install” the model; you configure access (provider + model + key) and your platform talks to their API.