What Can AI Do in Education? 12 Models and Platforms Compared
Miskola methodology · Updated 2026
Which AI models are best for education? Miskola’s comparison of 12 AI models and platforms helps teachers choose the right tool: each model’s strengths, limits and best classroom use cases. The comparison follows the Three-Level Architecture — matching models to Instrument, Mirror or Partner level tasks. Research-based guide by Gatis Šeršņevs (Miskola, Latvia).
What can be achieved when 7 different AI models and 2 platforms are given the exact same detailed pedagogical task? In this experiment, I compared 12 approaches, giving each an identical instruction (a single prompt, without corrections): to create professional methodological material on the topic “Linear Function for 8th Grade” in accordance with the Latvian Skola2030 standard.
The results are surprisingly different—each model has its own “handwriting,” strengths, and weaknesses. Some stand out with visual design, others with methodological depth, and still others with practical applicability. Below is the complete comparison with all 12 results, which you can open and explore yourself.
🎥 Video Review
📋 Testing Task (Prompt)
All 12 models/platforms received the same detailed instruction—”AI Educator: Professional Learning Process Development.” The prompt includes: 8 expert roles, 10 development steps, 11 mandatory output sections, and self-assessment criteria.
📄 Open the full prompt
🤖 Tested Models and Platforms
The table below summarizes all 12 results. Each model received identical instructions and created its own HTML document. Click on the model name to open its result.
| Model | Size | Characteristics |
|---|---|---|
| Qwen | 82.9 KB | Visually richest — interactive design with grid background, Space Grotesk fonts, sliders |
| ChatGPT | 36.0 KB | Simple, structured, academic — like a textbook chapter |
| Claude | 71.1 KB | Typographically sophisticated — Fraunces + Inter fonts, color-coded sections, elegant infographics |
| Gemini | 33.4 KB | Methodologically most complete — 5-hour lesson plan, Skola2030, UDL, assessment rubric, 10-error analysis |
| Grok 4.5 | 31.0 KB | Practical, with error analysis and teacher notes |
| Kimi 3 | 36.3 KB | Academic with dark cover design and teacher error analysis |
| GLM-5 | 97.0 KB | Most voluminous — 1736 lines, fixed sidebar, Font Awesome icons |
| Google AI Studio | — | Interactive material created in the Google AI Studio environment with instant publishing |
| Fable 5 | 35.5 KB | Anthropic Fable — clean, elegant design (Georgia), 12 sections with detailed square-approach representation, 12-error catalog, AI class contract |
| Opus 5 | 55.2 KB | Academically most developed — 12 sections, professional typography, detailed differentiation for 3 groups, 12-error analysis, full test |
| ChatGPT 5.6 Soul | 35.4 KB | The only one that created a full web application — separate HTML, CSS, and JS files. Interactive dark design |
| LM Arena | — | Result generated on the Chatbot Arena platform — comparison test |
🖼️
🔬
🔬 What do the results look like? An analytical comparison
Not just “which one is prettier”—but what it means for a teacher’s daily routine. Each model has its own strengths, and they only become apparent through comparison.
Qwen: Interactivity > Static Text
Qwen — the only one with interactive sliders
✅ What worked: The only model that stepped outside the framework of a “static HTML document.” Sliders for changing k and b values, buttons for selecting different functions — this is a different level of thinking about how a student will interact with the material.
📚 What it means for the lesson: Students can experiment — “what happens if I change k?” Without additional preparation. Inquiry-based activity is built right into the material itself.
❌ Drawback: Methodological content is superficial. Beautiful packaging, but less depth.
Gemini: Methodological Depth > Visual Brilliance
Gemini — the most complete methodological structure
✅ What worked: The only one that fully completed all 10 task sections. 5-hour lesson plan in an 8-column table (objective, learning outcomes, activities, teacher/student actions, resources, time). Direct connection to Skola2030. UDL across 3 levels. 10-error analysis with solutions.
📚 What it means for the lesson: This material can be used directly tomorrow morning. This is what a teacher wants to receive on a Friday evening before Monday.
❌ Drawback: No interactivity. Clean, structured text — the student has to imagine the visualizations themselves.
Claude: Design Thinking > Functionality
Claude — visually and structurally sophisticated
✅ What worked: Fraunces + Inter fonts, color-coded sections (blue/ochre/green), elegant icon infographics. Structure — clear hierarchy, easy to navigate.
📚 What it means for the lesson: Material you actually want to read. If you are creating handouts for students — the Claude version will look professional.
❌ Drawback: In places, more attention is paid to form than content. Lacks the methodological detail level of Gemini.
📊 Which model is good for what?
| Criterion | 🏆 Winner | Note |
|---|---|---|
| Ready to use in class | Gemini | Full 5-hour cycle |
| Interactivity | Qwen | Sliders, buttons |
| Visual design | Claude | Professional typography |
| Teacher’s guide | Gemini | Concrete tips |
| Differentiation | Gemini | 3 levels with justification |
| Error analysis | Gemini / Grok | 10 errors + solutions |
| Technical execution | ChatGPT 5.6 Soul | The only web application |
Conclusion: There is no single “best” model. Want to go to class tomorrow? Take Gemini. Want interactive material for students? Take Qwen. Want beautiful handouts? Take Claude. In an ideal world — combine the strengths of all three.
Author: Gatis Šeršņevs · Miskola (SIA Laba satura skola) · Latvia · Updated: August 2026