François Chollet noted that Argon's performance profile on the benchmark is similar to Astra's, indicating that Google has returned to the AI frontier. Other observers attribute the jump in scores to reinforcement learning on games, with one analyst estimating an 85% probability that RL drove the result compared to a 15% chance that 'loopers' were responsible, citing similar trends in the Opus 5 data.
Tuesday, October 20, 2026
Story
Google Gemini 4 Argon Hits 97.5% to Become Second Lab to Top ARC-AGI-3
AI5 days ago
A reasoning model from Google DeepMind has surpassed the 95% accuracy barrier on a test for artificial general intelligence, achieving a score of 97.5% using a provider adapter harness that cost $4.8K. The Gemini 4 Argon model scored 49.8% on the standard harness at a cost of $5.8K, while also recording 91.3% on the ARC-AGI-2 benchmark at $0.32 per task and 98.5% on ARC-AGI-1 at $0.16 per task. This performance makes the company the second frontier AI lab to exceed 95% on the ARC-AGI-3 benchmark.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.