ARC-AGI-3 saturated by Astra in 6 months, twice as fast as Chollet expected

ARC-AGI-3 发布仅半年即被 Astra 饱和，进展快于 François Chollet 预期一倍

Sherwin Wu says ARC-AGI-3, which he once found hard, is now saturated by Astra. François Chollet expected frontier models to take about a year; it took 6 months. The post doesn't disclose Astra's exact score or test details, so I'd hold off until full results land.

Why it matters: Saturating ARC-AGI-3 in 6 months vs. the expected 12 is a strong signal that directly updates priors on reasoning progress. The gap: no specific score or test conditions disclosed, so the claim gets a 30% discount. If a full report drops, this could hit 85.