한국어EN
Evaluation & BenchmarksARC Prize

ARC Prize releases ARC-AGI-3

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Released ARC-AGI-3 — humans 100% versus frontier AI 0.51%

Why it matters

The axis moved from static problem solving to interactive exploration and skill acquisition without instructions

Source — Primary sources first; secondary reports are labelled “Press”

Officialarcprize.orghttps://arcprize.org/blog/arc-agi-3-launch

Added in Week 35, 2026 · See that week

On the map — ⑯ Evaluation line

  1. Previous stopWhy SWE-bench Verified no longer measures frontier coding capabilities
  2. This record · stopARC-AGI-3
  3. Next stopRealSWE
See it on the map

Same week

YOLO-World v2 applied to MVTec AD with an extreme 1:9 train-validation splitOfficialSakana AI's AI Scientist research is published in NatureOfficialCohere releases Cohere TranscribeOfficialHarness design for long-running application developmentOfficial

Nearby on the same axis

Why SWE-bench Verified no longer measures frontier coding capabilitiesOfficialLaude Institute releases Terminal-Bench 2.1OfficialMETR releases Time Horizon 1.1OfficialLMArena renames itself ArenaOfficial