Google's Gemini 4 Argon hits 97.5% on ARC-AGI-3 benchmark

google deepmind's new frontier model, gemini 4 argon, scored 97.5% on the arc-agi-3 benchmark at a cost of $4,800 using the provider adapter harness. this makes google the second ai lab to cross the 95% mark on this benchmark, following openai's gpt-6 astra which scored 99.9%. argon's performance is notable, requiring 62% fewer actions than the human baseline across 25 public games. however, its performance varies by domain, with lower scores in math and software engineering. in the standard harness, argon's score drops to 49.8% at low reasoning for $5,800.