Is AI Smarter Than You? I Tested 60 Years of It Against Real Students

Ask how smart AI is, and you'll find IQ numbers everywhere. Most of them are guesses. So I built a fresh test from published research and gave the same 43 questions to 60 years of AI: a thermostat rule, ELIZA from 1966, a Deep Blue-strength chess engine, GPT-2, two small chatbots and ten runs of today's models. Every rung of the ladder is set by how real students did. Results (42 puzzles, 7 rungs): · GPT-6 Astra (max) and Gemini 4 Argon (high): 42/42 · Claude Opus 5.5 (max) 39 · Kimi K3.1 Pro (max) 38 · Claude Haiku 5.5, GPT-6 Luna and Opus 5.5 (low) 37 · GPT-6.1 Sol and Claude Haiku 4.5 36 · Claude Sonnet 5.5 33 · Every recent model climbed all seven rungs. GPT-6 Luna did it for an estimated API cost under one cent. · GPT-2 XL (2019) 11 · Qwen 2.5 0.5B (2024) 3 · TinyLlama (2023) 1 · ELIZA 0 · thermostat 0. The chess engine could only read the chess puzzle, and solved it. How we tested: · 42 new puzzles made with the generators of Webb, Holyoak & Lu (Nature Human Behaviour, 2023), six per rung, plus one Lichess mate-in-two. · A system climbs a rung if it matches the students' share on that puzzle type (85% on rung 1, 28% on rung 7), and it can't skip one. This isn't an IQ test: it's a student baseline on the same kinds of puzzle. · Questions, answer key and scoring rules were locked with SHA-256 before the scored runs. Today's models answered in their own official tools, no tools allowed, one page of six puzzles at a time. Chapters: 0:00 What is the IQ of an AI? 0:34 The test 0:55 Try a puzzle 1:22 A ladder set by real students 2:28 A thermostat 2:46 ELIZA, 1966 3:27 A Deep Blue-strength chess engine 3:55 GPT-2, 2019 5:12 Two small chatbots 5:49 The jump 6:11 Today's models 6:40 The puzzle they split on 7:15 Effort and price 7:48 The ladder ran out 8:07 IQ 151 on TrackingAI 8:41 The unpublished test: 136 9:12 What an IQ number really means 9:40 ARC-AGI-3 10:16 METR: 3 seconds to 17 hours 10:55 1,200 agents went rogue 12:37 Red lines 13:38 So, what is the IQ of AI? 14:07 ELIZA comes back Take the test yourself. The questions, the answer key, every model's raw answers and a script that re-scores them: https://github.com/novaki-ai/ai-iq-la... The exact prompt is in the pinned comment. Sources: · Webb, Holyoak & Lu: https://arxiv.org/abs/2212.09196 · TrackingAI (Maxim Lott): https://trackingai.org/IQ · Mensa Norway test: https://test.mensa.no · ARC-AGI-3 leaderboard: https://arcprize.org/leaderboard · METR time horizons: https://metr.org/time-horizons/ · OpenAI on the Hugging Face incident: https://openai.com/index/hugging-face... · Hugging Face: https://huggingface.co/blog/security-... · METR's investigation: https://metr.org/blog/2026-08-26-open... · GPT-6 Astra system card: https://deploymentsafety.openai.com/g... · Anthropic, Project Glasswing: https://www.anthropic.com/glasswing · IDAIS Beijing red lines: https://idais.ai/dialogue/idais-beijing/ · Global Call for AI Red Lines: https://red-lines.ai/ · ELIZA and GPT-4o in a Turing test: https://arxiv.org/abs/2503.23674 Subscribe:    / @novaki_ai   The narration uses an AI voice (Google Gemini). Website: https://novaki.lol · X: https://x.com/Novaki_ai · Instagram:   / novaki_ai   Contact: hello@novaki.lol #AI #IQ #ArtificialIntelligence