See how different AI models respond to each GIF CAPTCHA — and why they succeed or fail
Choose a model to see its simulated responses to all 10 GIF CAPTCHAs, along with capability analysis explaining why it passes or fails each one.
Click each card to expand the full response, human comparison, capability breakdown, and reasoning.
Pass/fail matrix across all models and CAPTCHAs. Green = AI fails to solve (CAPTCHA works), Red = AI solves it (CAPTCHA broken).
How many CAPTCHAs block each model (higher = more effective)
Average capability scores across all CAPTCHA dimensions
GIF #6 (Parent Dog Sacrifice) and #9 (Road Rage Hug) remain hardest for all models. They require emotional comprehension and social script violation detection — capabilities that even 2025 multimodal models struggle with.
GIF #3 (Flying Skateboarder) and #8 (Highway Drift) are solvable by modern vision models. They rely on motion tracking — a capability that multimodal models with frame analysis can now handle.
Responses are simulated based on each model's documented capabilities (text-only vs. multimodal, frame analysis vs. video understanding, cultural training data). Results reflect estimated performance, not actual API calls. The simulation captures the type of response each model would generate — from complete inability (GPT-4 2023) to partial frame-based inference (GPT-4o) to emerging video comprehension (Gemini 1.5 Pro).