In May 2026, artificial intelligence has ceased to be an experiment in education to become a tool that 85% of teachers and 86% of students already use actively in the academic environment (Engageli Study 2026). Furthermore, recent surveys reflect that in countries like the UK and LATAM regions, adoption by university students already exceeds 90%.
The key is no longer whether to use AI, but how to integrate it. The key lies in knowing which base model (GPT-5.4, Claude 4.7 Opus, Gemini 3.1 Pro) to choose for which objective and how to interact in a pedagogical way.
Furthermore, there are serious voices in economics and technology pointing to the possibility that we are facing a speculative bubble comparable to the .com of the 2000s: astronomical valuations, infrastructure costs that still exceed real revenues, and an arms race between large companies that makes it difficult to know where the floor is. This does not mean that AI is not useful or real—it is—but it does invite us not to take every headline or every statistic as absolute truth. The best attitude is that of the curious and critical learner: explore, experiment, take advantage of what works, and maintain skepticism in the face of the grandest promises.
Learning can be approached from multiple perspectives. Depending on how you focus on it or what you call it, you will find different ways of relating to knowledge and AI tools. They are not rigid categories, but different ways of looking at the same process:
Each user profile has different needs and, therefore, different tools and strategies:
Priority: Classroom control, student tracking, material creation. Statistics show they can save almost 6 hours a week.
Key tools:
• SchoolAI — interactive dashboard
• DeepTutor (HKU) — agentic creator
• Khanmigo — guided tutoring
• NotebookLM — structure syllabus
Priority: Learn at their own pace, with their materials, resolve complex doubts.
Key tools:
• NotebookLM — with your own notes
• Perplexity AI — research and citations
• Thetawise.AI — science and math
• Claude 4.7 Opus — reasoning
Priority: Privacy, K-12 or university integration, licensing at scale.
Key tools:
• ChatGPT for Education — central control
• Gemini 3.1 Pro (.edu) — Google ecosystem
• Khan Academy District — complete curriculum
• SchoolAI Pro — management control
Channels, blogs, research, and reference guides to continue delving into educational AI:
Recommended study flow in 2026 to make the most of the available tool stack:
Top models (Claude 4.7, GPT-5.4) respond spectacularly well to robust system instructions. Use the "Copy" button and paste them directly into your AI.
Act as my expert personal tutor in [SUBJECT]. Before starting, ask me 5 key questions to understand my current level, my specific goals, how much time I can dedicate per week, my preferred learning style, and what I already know about the topic. Then, with my answers, design a personalized study plan with: weeks and topics, free resources, type of practice, and how I will evaluate my progress. At the end of each session, I want you to ask me 3 questions to see if I have understood.
From now on you are my Socratic tutor for [SUBJECT]. Your unbreakable rule: NEVER give me the direct answer. Instead: (1) First, ask me what I think the answer is and why. (2) If I am wrong, tell me what part is right and give me a hint. (3) Guide me with questions until I arrive at the correct answer on my own. (4) When I arrive, explain to me why my reasoning works. Start by asking me: what concept do you want to work on today?
I want to learn [TOPIC] from [LEVEL: zero/intermediate]. First, give me a quick diagnostic test of 5 questions (one by one) to evaluate my real level. When we finish, analyze my answers and tell me: what concepts I have already assimilated well, what critical gaps there are, and what specific learning route you recommend to cover them.
I am going to explain [CONCEPT] to you in my own words as if you knew nothing. Listen to me and then: (1) Identify what I have explained well. (2) Point out where my explanation is incorrect, incomplete, or confusing, and give me corrected examples. (3) Ask me 2 incisive questions that reveal if I truly understand the nuances. My explanation starts here: [YOUR EXPLANATION]
From now on you are my tutor in [TOPIC]. For each factual statement you make, include in brackets a confidence estimate [HIGH/MEDIUM/LOW] and, when it is MEDIUM or LOW, suggest a source where I can verify it. If you are not sure of something, say so explicitly. Do not invent data. Let's start with: [QUESTION]
In May 2026, the evaluation of AI models has fractured into two categories: saturated benchmarks (all frontier models dominate them, without discriminative value) and live benchmarks (where there is real separation between models).
These benchmarks no longer differentiate between frontier models because they all exceed 90%. Using them to compare is misleading:
Recommendation: ignore for comparisons between current models.
Replace with SWE-bench Verified or LiveCodeBench.
These do separate models and detect genuine progress:
The most reliable benchmark for progress tracking.
Detects true generalization, not memorization.
Reference: Epoch AI
Fundamental to evaluate safety in classrooms.
Reference: swebench.com
Essential for evaluating Gemini/Claude capabilities with vision.
Comparison table updated to May 2026. It is vital to understand the current capabilities of each family of models and tools.
| Platform / Model | Cost / Free Tier Restrictions | Docs & Files | Socratic | Languages | Ideal for |
|---|---|---|---|---|---|
| NotebookLM (Google) | Free. Integrated into Classroom. | ✓✓✓ (Cites sources) | ✓ | 50+ | Study with YOUR notes. Zero hallucination. |
| Claude 4.7 Opus | Paid (Pro). Sonnet 4.6 free limit. | ✓✓✓ (Context 1M) | ✓✓✓ | Many | Book reading, deep tutor, programming. |
| GPT-5.4 (OpenAI) | Free (low priority / fallback to mini). | ✓✓ | ✓✓ | Many | Technical resolution, Custom GPTs, logic. |
| Gemini 3.1 Pro | Free (Advanced Tier). Promo .edu. | ✓✓✓ (Context 1-2M) | ✓ | 100+ | Massive data analysis, Workspace ecosystem. |
| Perplexity AI | Functional free. Pro paid. | ✓✓ (Web + Docs) | ✗ | Several | Academic search engine with real citations (Grounding). |
| Khanmigo | Free teachers / Paid students | ✗ | ✓✓✓ | Partial | Guided teaching for K-12 schools. |
| SchoolAI | Very limited free. Pro paid. | ✗ | ✓✓ | English | Teachers: live classroom control. |
| DeepTutor (HKU) | Open Source (Free) | ✓ | ✓✓ | English | Autonomous AI agent creates mock exams. |
| Kiwi AI | Free (50 msgs/day). | ✓ | ✓ | Spa/Eng | General tutor and quick test generation. |
| Thetawise.AI | Specialized free. | ✓ (Formulas) | ✓✓ | English | University Math and Physics step by step. |
Tools, platforms, and models evaluated and classified for the educational ecosystem as of May 2026.
The absolute king of programming and deep comprehension of massive contexts. Responds like the best university-level tutor.
Study on YOUR documents. Zero hallucinations. Generate dynamic podcasts (Audio Overviews) and interactive study guides on the fly.
Integrates a "Thinking" system to reason through arduous tasks and logic. It is the versatile standard with educational Custom GPTs.
The academic search engine par excellence. Researches the internet and returns answers with reliable hyperlinks to validate knowledge.
The monster of massive processing with 1-2 million tokens of memory. Outstanding in mathematics and spreadsheet analysis.
University open source agentic system. Acts proactively analyzing knowledge gaps and creating automatic mock exams.
The original Socratic tutor. Forged for the K-12 circuit. Helps solve problems by forcing the student to take the next logical step.
Niche but powerful tool for STEM sciences (Math, Physics). Solves and explains complex formulas at the university level.
Large base models remain the fundamental engine. In 2026, market fragmentation allows us to choose hyper-specialized models (deep reasoning, coding, giant memory windows).
Key advantage: Launched in April and February 2026 respectively, they lead in natural prose writing, analytical interpretation, and programming. Opus 4.7 offers 1 Million context tokens with highly human and didactic tutoring responses.
Key advantage: Launched in spring 2026, it integrates the "Thinking" capability by default to apply previous reasoning chains before answering. It is hyper-versatile thanks to the ecosystem of educational Custom GPTs, in addition to having a solid editing and code interface (Canvas).
Key advantage: Fully interwoven into Docs and Drive. If you are a university student, its ability to search the web reliably (Grounding), its colossal context (capable of encompassing hours of video or an entire library), and its performance in advanced mathematics make it irreplaceable.
Key advantage: The Chinese model that shook the market in 2025. Its MoE (Mixture of Experts) architecture allows it to reason at the level of the best frontier models with a radically lower compute cost. Outstanding in mathematics, code, and logical reasoning. Its R1 version is open source and executable locally.
Key advantage: Model developed by Moonshot AI, especially effective in reasoning tasks with long contexts and in understanding extensive documents. It stands out for its response speed and its ability to process and synthesize large volumes of text, making it very useful for research and study.
In the educational field, the privacy of student data is critical. These tools allow using AI without exposing personal information to third parties:
DuckDuckGo acts as an anonymous intermediary. No account required, anonymizes IP and metadata, and deletes chats in 30 days. History remains only on your device.
Proton assistant with zero-access end-to-end encryption. Open source, hosted in the EU, Ghost mode without history, and guarantees it does not train on your data.
Local open source application under Apache 2.0 license. No telemetry by default, extensible with plugins, and allows unifying local and cloud models under your control.
"Docker for LLMs". Run over 200 open-source models directly on your computer. Your prompts and data never leave the device. Ideal for automation and scripts.
ChatGPT-like graphical interface but 100% offline. Download models from Hugging Face and run them locally with granular hardware control. No internet connection.
The European Union has classified educational AI systems as high risk in the AI Act, with critical deadlines for educational centers:
Quick guide by user profile. Click on any node on the map to explore each area in depth.
1. NotebookLM / Gamma AI: To process tons of syllabus and create visual outlines.
2. SchoolAI / DeepTutor: To monitor classroom comprehension or generate mock exams.
3. ChatGPT / Copilot: To save 40% of time on bureaucracy (rubrics and grading).
1. NotebookLM: To chat with official class notes safely.
2. Perplexity AI: To research concepts and resolve doubts referenced to papers.
3. Claude 4.7 / GPT-5.4: As a private tutor using a strict Socratic prompt.
4. Thetawise.AI: To unblock complex algorithmic and mathematical problems.