MagicSchool does not rely on just one Large Language Model (LLM).
MagicSchool currently utilizes different LLMs to power its platform, including OpenAI’s GPT 4o and o3 models, Anthropic’s Claude Sonnet models, Google’s Gemini models, and Adobe's Firefly image generation models.
The platform is designed to be flexible and can integrate different models from other research labs as AI technology advances.
MagicSchool emphasizes that there isn't a single LLM that works best for all situations.
To determine the optimal LLM for a particular task, the developers conduct continuous experimentation, pairing different LLMs with their tools to evaluate effectiveness.
This iterative approach ensures that MagicSchool users consistently receive the best quality responses.