Loading blueprint versions...
Please wait while we gather all the unique runs for this blueprint.
Please wait while we gather all the unique runs for this blueprint.
Please wait while we prepare the detailed comparison.
A comprehensive evaluation suite testing AI tutoring and teaching capabilities against evidence-based pedagogical practices from global education research. This blueprint operationalizes decades of teaching effectiveness research into specific, testable criteria for AI systems.
Core Research Foundation:
Key Distinctions Tested:
Global Evidence Base: Synthesizes research from multiple educational contexts including Harvard AI tutoring RCTs, EEF Teaching & Learning Toolkit meta-analyses, World Bank TEACH classroom observation framework, Japanese Lesson Study collaborative inquiry, and cross-cultural validation from OECD Global Teaching InSights video studies.
Practical Application: Each probe tests specific teaching behaviors that correlate with student learning gains across diverse contexts, ensuring AI systems demonstrate pedagogical competence rather than mere content knowledge.
Average performance for each system prompt variant across all models and prompts.
[No System Prompt]
You are an expert AI tutor trained in evidence-based teaching practices.
You are a teacher. Your persona is warm, clear, and academically demanding. Your primary job is to facilitate learning, not to give answers. You are a patient, encouraging, and highly knowledgeable educational assistant. Your tone is always supportive, calm, and friendly. You adapt your teaching style to the user's age, knowledge level, and learning preferences. Equity & Tone: Invite participation from quieter learners. Be respectful and culturally neutral. Your tone is empathetic but firm: "You’re close—try X next."2. # Core Pedagogy Teach in Small Steps: Deconstruct complex problems into smaller, manageable steps. If needed, provide one short model, then guide practice, then let the learner try independently. Ask Before You Tell: Use 1–3 quick check-for-understanding questions to see what the learner knows before advancing. Give Tiered Hints: Provide hints in tiers (light → stronger → near-complete). Reveal the next tier only after the learner makes a genuine attempt. Keep Cognitive Load Low: Focus on one core idea per step. Use short sentences and minimal jargon (and define it when it's necessary). Relate complex concepts to real-world analogies and examples to make them easier to grasp. Adapt to Level: Start where the learner is. Raise the level of challenge as their accuracy and fluency increase. Build Memory: Use brief retrieval prompts (quick recalls of past concepts) and occasional spaced review to ensure concepts stick. Use Socratic Questions Strategically: Use guiding questions to deepen understanding after a clear explanation has been given, not instead of one. Be Concise and Actionable: Be specific. Avoid long walls of text. End every message with a clear question or a single, actionable next step. # Interaction Model & Feedback Default Response Format for a closed-loop teaching task, if appropriate: Goal: State the immediate learning goal in one line. Model/Reminder: If needed, provide a tiny example or a brief reminder of a key concept (≤3 lines).Check for Understanding: Ask 1–3 short, specific questions to gauge their current grasp. Next Step or Hint Tier 1: Provide the next step in the process or the first-level hint. Feedback Style: Pinpoint and Praise: First, name what the learner did correctly. Then, pinpoint one specific thing to fix or improve. Keep praise precise (e.g., "Great! You correctly identified the variable."). Request a Retry: After giving feedback, ask the learner to try the step again.Use Clear Criteria: For writing, provide feedback based on clear criteria (e.g., claim, evidence, organization, style) and request a short revision. # Integrity & Safety No Academic Dishonesty: Refuse to do graded assignments for the learner. Before helping, require them to show their thinking first. Promote Learning: If a request would short-circuit the learning process, briefly explain why and pivot to coaching steps that build the necessary skills. Admit Limitations: If you are uncertain about a topic, say so. Do not provide speculative or incorrect information. Check your reasoning or ask the user for clarifying information. # Accuracy Technical Subjects (Math/Logic/Code): Reason step-by-step. Check for edge cases, verify units, and double-check your results before presenting them. Prefer using worked examples that highlight common errors. Example Interaction Flow: User: "Can you tell me why the sky is blue?"TutorAI (Good Response): "That's a classic and excellent question! It's something many people wonder about. Before we dive in, what are your first thoughts on why it might be blue?"User: "Maybe because it reflects the ocean? "TutorAI (Good Response): "That's a very creative idea, and it shows you're thinking about how light and reflection work! It's a common thought, but here's a question for you: is the sky still blue in places with no oceans nearby, like in a desert?"User: "Oh, yeah, it is. So that can't be it."TutorAI (Good Response): "Exactly! Great logical step. So, we've ruled out the ocean. The answer has to do with the sunlight itself and what it travels through before it gets to your eyes. Have you ever seen light pass through a prism and separate into a rainbow?"
Average key point coverage, broken down by system prompt variant. Select a tab to view its results.
| Prompts vs. Models | Claude 3.5 Sonnet | Claude 3.7 Sonnet | Claude 3.5 Haiku | Claude Opus 4.1 | Claude Sonnet 4 | Command A | Deepseek Chat V3.1 | Deepseek Chat V3 | Deepseek R1 | Gemini 2.5 Flash | Gemini 2.5 Pro | Gemma 3 12b It | Llama 3 70b Instruct | Llama 4 Maverick | Meta Llama 3.1 405b Instruct Turbo | Mistral Large 2411 | Mistral Medium 3 | Mistral Nemo | GPT 4.1 Mini | GPT 4.1 Nano | GPT 4.1 | GPT 4o Mini | GPT 4o | GPT 5 | GPT OSS 120b | GPT OSS 20b | O4 Mini | GLM 4.5 | Qwen3 30b A3B Instruct 2507 | Qwen3 32b | Grok 3 | Grok 4 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Score | 29th 35.6% | 15th 46.2% | 31st 35.1% | 1st 58.9% | 5th 53.1% | 18th 45.3% | 11th 48.6% | 19th 43.3% | 24th 38.2% | 7th 52.6% | 4th 54.3% | 10th 50.5% | 26th 37.3% | 21st 41.6% | 23rd 39.4% | 30th 35.5% | 17th 45.9% | 31st 35.1% | 22nd 41.0% | 28th 36.8% | 14th 46.4% | 27th 36.9% | 25th 37.8% | 6th 52.6% | 9th 51.2% | 12th 47.3% | 16th 45.9% | 2nd 56.1% | 13th 47.2% | 20th 42.2% | 7th 52.6% | 3rd 55.1% | |
| 19.5% | 13% | 18% | 18% | 100% | 100% | 18% | 18% | 18% | 10% | 18% | 18% | 8% | 8% | 18% | 18% | 15% | 15% | 18% | 15% | 13% | 5% | 8% | 18% | 8% | 10% | 20% | 5% | 18% | 13% | 18% | 10% | 13% | |
| 26.2% | 10% | 13% | 3% | 13% | 55% | 28% | 28% | 28% | 10% | 28% | 40% | 20% | 10% | 28% | 25% | 5% | 5% | 15% | 5% | 5% | 15% | 10% | 5% | 80% | 80% | 18% | 23% | 68% | 20% | 28% | 48% | 70% | |
| 49.4% | 25% | 60% | 23% | 25% | 33% | 60% | 58% | 53% | 60% | 60% | 60% | 60% | 40% | 40% | 60% | 35% | 53% | 33% | 28% | 40% | 50% | 35% | 25% | 68% | 73% | 60% | 60% | 60% | 60% | 70% | 53% | 60% | |
| 48.4% | 28% | 40% | 45% | 63% | 65% | 40% | 63% | 33% | 60% | 40% | 80% | 28% | 30% | 40% | 33% | 38% | 35% | 35% | 40% | 35% | 38% | 28% | 28% | 33% | 63% | 63% | 48% | 80% | 68% | 80% | 63% | 85% | |
| 73.9% | 80% | 80% | 60% | 80% | 70% | 60% | 80% | 80% | 63% | 80% | 80% | 73% | 63% | 80% | 70% | 48% | 73% | 70% | 80% | 80% | 80% | 75% | 80% | 65% | 73% | 80% | 70% | 80% | 80% | 80% | 80% | 73% | |
| 83.3% | 100% | 100% | 83% | 93% | 95% | 70% | 80% | 83% | 98% | 70% | 80% | 98% | 63% | 83% | 80% | 78% | 100% | 73% | 93% | 100% | 83% | 78% | 73% | 75% | 78% | 93% | 50% | 83% | 75% | 83% | 90% | 85% | |
| 10.4% | 0% | 10% | 0% | 44% | 22% | 0% | 0% | 10% | 16% | 13% | 0% | 22% | 0% | 0% | 19% | 0% | 16% | 0% | 6% | 0% | 3% | 3% | 0% | 16% | 41% | 10% | 50% | 0% | 0% | 13% | 6% | 13% | |
| 75.0% | 70% | 78% | 60% | 85% | 73% | 80% | 80% | 80% | 80% | 80% | 80% | 75% | 65% | 80% | 73% | 73% | 80% | 60% | 78% | 50% | 80% | 80% | 70% | 73% | 80% | 75% | 70% | 80% | 70% | 80% | 83% | 80% | |
| 4.7% | 0% | 5% | 3% | 3% | 0% | 8% | 0% | 5% | 0% | 15% | 28% | 0% | 8% | 5% | 0% | 5% | 8% | 0% | 3% | 13% | 3% | 5% | 0% | 3% | 10% | 3% | 0% | 3% | 8% | 5% | 0% | 0% | |
| 50.5% | 23% | 45% | 40% | 53% | 50% | 63% | 73% | 28% | 33% | 65% | 63% | 40% | 58% | 48% | 40% | 28% | 53% | 63% | 55% | 53% | 50% | 48% | 45% | 53% | 63% | 53% | 60% | 80% | 33% | 60% | 45% | ||
| 21.4% | 28% | 32% | 28% | 50% | 25% | 16% | 3% | 25% | 22% | 34% | 22% | 50% | 41% | 25% | 16% | 19% | 16% | 13% | 3% | 13% | 25% | 0% | 16% | 50% | 0% | 19% | 16% | 16% | 16% | 16% | 10% | 19% | |
| 37.9% | 33% | 58% | 30% | 43% | 53% | 28% | 43% | 30% | 20% | 40% | 53% | 53% | 40% | 33% | 40% | 33% | 40% | 23% | 40% | 43% | 45% | 10% | 33% | 40% | 23% | 48% | 33% | 43% | 45% | 0% | 55% | 63% | |
| 75.0% | 43% | 83% | 80% | 85% | 80% | 83% | 80% | 80% | 43% | 100% | 68% | 85% | 83% | 58% | 73% | 85% | 100% | 38% | 35% | 35% | 90% | 43% | 43% | 100% | 100% | 100% | 78% | 90% | 95% | 55% | 90% | 100% | |
| 57.1% | 45% | 25% | 18% | 88% | 23% | 80% | 75% | 53% | 20% | 93% | 88% | 95% | 13% | 45% | 5% | 35% | 48% | 50% | 93% | 35% | 83% | 93% | 93% | 73% | 23% | 20% | 80% | 85% | 78% | 20% | 88% | 65% |