Chollet: 'ARC-AGI-3 Is the Only Unsaturated Agentic AI Benchmark' — GPT-5.4 Scores 0.26%, Opus 0.25%, Gemini 0.37%

X/Twittervia @fchollet·high signal

François Chollet tweeted that ARC-AGI-3 is now the only unsaturated agentic AI benchmark, with sub-1% scores from all frontier models on the private test set. Specific scores: Gemini 3.1 Pro 0.37%, GPT-5.4 High 0.26%, Opus 4.6 0.25%, Grok-4.20 0.00% — while humans score 100%. Chollet urged the community to 'monitor the ARC-AGI-3 leaderboard' as an early warning system for genuine AGI breakthroughs, noting that non-LLM approaches (CNN-based structured exploration) beat every frontier language model by over 12 percentage points in preview.