Is ARC-AGI-2 still meaningful?

Known problems

• It went from 50% to 90% in seven months.^{[2][2] Benchmark data bundle Epoch AI Primary}
• Anthropic dropped ARC-AGI from its Claude Opus 5.5 launch table (22 Sep 2026).^{[3][3] Introducing Claude Opus 5.5 Anthropic Vendor}

What to look at instead

• ARC-AGI-3 Status:Contested Yes, but only with the harness named. One adapter moves the top score by 37 points.

Where it appears in the report

Chapter 1, Benchmarks

• Figure 1. Each New Benchmark Fills Up Faster Than the Last
• Figure 2. Recent benchmarks went from 50% to 90% in under a year
• Table 2. First model to cross each threshold in “Go deeper: The full saturation timeline (14 benchmarks)”
• Table 3. How each area replaced its saturated benchmarks, oldest first in “Go deeper: Lineage of replacements, by area”
• Table 5. Reasoning and knowledge benchmarks: what they measure, who maintains them, the top score on 5 November 2026 and the known flaw in “Go deeper: Reasoning and knowledge benchmarks (17)”