Development

A new benchmark returns frontier models to near zero

development2026-03-27

Claim

ARC-AGI-3 launched with 135 novel game environments designed to be trivial for humans and humbling for machines, with Gemini 3.1 Pro scoring 0.37%, GPT-5.4 0.26%, Opus 4.6 0.25% and Grok 4.2 zero.

Reported metric: 0.37% best

Reporting date: 2026-03-27 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to March 27, 2026 (theinnermostloop.substack.com)

The first benchmark in months to open a gap rather than close one.

Knowledge graphopen in browser →

About

ARC-AGI-3

Reported in

Welcome to March 27, 2026

Actor

ARC Prize Foundation

Evidence for

Supersedes

A conjecture from 2019 is settled and its author approves

Referenced by

Cite or correct this record

Citation formatCitation text

A new benchmark returns frontier models to near zero
Innermost Loop KB — secondary record
Reporting date: 2026-03-27
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-03-27-arc-agi-3-humbles-the-frontier
Claim: ARC-AGI-3 launched with 135 novel game environments designed to be trivial for humans and humbling for machines, with Gemini 3.1 Pro scoring 0.37%, GPT-5.4 0.26%, Opus 4.6 0.25% and Grok 4.2 zero.
Attributed sources:
- Welcome to March 27, 2026 — https://theinnermostloop.substack.com/p/welcome-to-march-27-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.