Google DeepMind’s Gemini 4 Argon Scores 97.5% on ARC-AGI 3 Benchmark

15 Oct 2026, 11:23 pm · 1h ago · 1 min read · OfficeChaiNew

Google DeepMind’s new Gemini 4 Argon model has recorded a 97.5% score on the ARC-AGI 3 evaluation benchmark when tested with harness scaffolding, placing it just behind OpenAI’s GPT-6 Astra. The ARC-AGI benchmark is widely recognized across the artificial intelligence sector as a rigorous test for measuring general reasoning and abstract problem-solving capabilities in frontier models. Gemini 4 Argon’s performance underscores substantial advancements in reasoning architectures among leading AI research labs. The result signals accelerating competition at the frontier of foundation models as labs push toward human-level adaptability in synthetic intelligence.