Noesion Radarnoesion.ai
Research2026-08-06

ARC Prize published verified test results for Google's Gemini 3.6 Flash across its four reasoning effort settings on the ARC-AGI-1 and ARC-AGI-2 abstract puzzle benchmarks. At the highest effort the model solved 91.2% of ARC-AGI-1 tasks and 60.4% of ARC-AGI-2 tasks, with per-task costs listed for each. Scores fall sharply at lower effort settings, dropping to 2.6% on ARC-AGI-2 at the minimal setting, and no ARC-AGI-3 results were reported.

What it unlocks

Comparing accuracy and cost per task across four reasoning effort settings before choosing one for a workload.

  • 91.2% on ARC-AGI-1 Semi-Private at high effort
  • 60.4% on ARC-AGI-2 Semi-Private at high effort
  • $0.34 per task on ARC-AGI-1, $0.61 per task on ARC-AGI-2
  • Minimal reasoning: 34.5% and 2.6%

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.