Development

A harness takes ARC-AGI-2 to 75% without touching the model

development2025-12-24

Claim

Poetiq's harness on GPT-5.2-xhigh reached 75% on ARC-AGI-2 at under $8 per problem, beating the prior state of the art by about 15% with no modification to the base model.

Reported metric: 75%, <$8/problem

Reporting date: 2025-12-24 (not necessarily the event date)

Sources

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

Welcome to December 24, 2025 (theinnermostloop.substack.com)

The scaffolding thesis producing the headline number, not a footnote.

Knowledge graphopen in browser →

About

Reported in

Welcome to December 24, 2025

Actor

Evidence for

Supersedes

Zoom's federated swarm beats the monolith on Humanity's Last Exam

Referenced by

Cite or correct this record

Citation formatCitation text

A harness takes ARC-AGI-2 to 75% without touching the model
Innermost Loop KB — secondary record
Reporting date: 2025-12-24
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2025-12-24-poetiq-harness-arc-agi-2
Claim: Poetiq's harness on GPT-5.2-xhigh reached 75% on ARC-AGI-2 at under $8 per problem, beating the prior state of the art by about 15% with no modification to the base model.
Attributed sources:
- Welcome to December 24, 2025 — https://theinnermostloop.substack.com/p/welcome-to-december-24-2025
This citation does not imply independent verification of linked-source claims.

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