Gemini 3.6 Flash achieves 91.2% on ARC-AGI-1 benchmark
August 6, 2026
Gemini 3.6 Flash reached 91.2% on ARC-AGI-1 and 60.4% on ARC-AGI-2, with task costs estimated at $0.34 and $0.61 respectively. In contrast, Gemini 3.5 Flash-Lite scored 53.5% on ARC-AGI-1 at $0.09 per task.
HOW THIS AFFECTS YOU
●
builderYou can evaluate these models for reasoning-heavy tasks using specific cost-per-task metrics.
●
researcherThe performance jump on ARC-AGI-2 suggests improved systematic generalization capabilities in the Flash architecture.