Tier 1
Frontier paid
GPT-4o, Claude Sonnet, Gemini 1.5 Pro
What you can do
Generate polished explanations, compare revisions quickly, and ask for multiple activity formats in one sitting.
What is lost vs. tier above
This is the top tier in this comparison; the loss is not capability but openness, cost transparency, and reproducibility.
Who is excluded
Participants without paid accounts, approved payment methods, stable broadband, compatible devices, or permission to use commercial AI systems.
Tier 2
Free / limited
GPT-3.5, Claude Haiku free tier, limited Gemini
What you can do
Run basic explanations, collect first drafts, and compare prompt changes when usage limits do not interrupt the lesson.
What is lost vs. tier above
Less reliable reasoning, shorter context windows, lower availability, fewer multimodal features, and unpredictable rate limits.
Who is excluded
Participants without accounts, phone verification, consistent internet, district approval, or enough time to retry blocked requests.
Tier 3
Unplugged / local
No AI - pen + paper, the class, local/offline models
What you can do
Model prediction by hand, inspect examples prepared ahead of time, or use offline systems where available.
What is lost vs. tier above
Live generation, novelty on demand, fast comparison across many prompt versions, and the feeling of direct interaction with a model.
Who is excluded
Fewer people are excluded by accounts or cost, but some learners may lose access to assistive features, translation, or individualized pacing.