MindPattern
← The Wire

Public story · 2026-02-24 · source-backed

Poetiq Architecture: How a 6-Person Team Beat Google on ARC-AGI-2

Confidence
source-backed
Sources
2
Redaction
passed

Story

poetiq.ai — Detailed technical breakdown of how a $40K-hardware startup achieved 54% on ARC-AGI-2 (vs. Google's 45% at nearly 3x the cost). The key innovation: "learned test-time reasoning" — an iterative refinement meta-system where solutions are generated, receive structured feedback, and self-improve through multiple loops. Crucially, this is LLM-agnostic and works across OpenAI, Anthropic, and xAI models. The meta-system IS the intelligence, not the underlying model. Symbolica's parallel work (85.28% with recursive delegation) confirms the pattern independently.

Related stories

Each link below shares sources, entities, or timing with this story.

  1. Shared entities / Same source / Shared topic

    Recursive Sub-Agent Delegation: The Winning Pattern for Complex Tasks

    Both cover AGI, Anthropic, LLM, OpenAI; cite the same source (poetiq.ai, Symbolica's parallel work); overlapping topics (anthropic, arc-agi-2, beat, delegation).

  2. ARC-AGI-2 Breaks 50% — Agent Scaffolding Beats Raw Model Power

    Both cover AGI, ARC, Google, Symbolica; cite the same source (poetiq.ai, Symbolica's parallel work); overlapping topics (arc-agi-2, architecture, beat, model).

  3. Shared entities / Shared topic / What happened next

    ARC-AGI-3 Humbles Every Frontier Model. A Simple CNN Beats Them All by 30x.

    Both cover AGI, Anthropic, ARC, LLM; overlapping topics (feedback, model); picks up the AGI thread on 2026-03-29.

  4. Shared entities / Shared topic / What happened next / Tension

    Google, Microsoft, and xAI Give US Government Pre-Launch Model Access.

    Both cover Anthropic, Google, OpenAI; overlapping topics (anthropic, google, model); picks up the Anthropic thread on 2026-05-10.

  5. Shared entities / What happened next / Tension

    An Open-Source Model Just Beat GPT-5.4 and Opus 4.6 on the Hardest Coding Benchmark

    Both cover Anthropic, Google, LLM, OpenAI; picks up the Anthropic thread on 2026-04-20; pushes against this story (against).

  6. Shared entities / Same source domain / Shared topic / What happened next

    Symbolica Hit 36% on ARC-AGI-3 for $1,005. Frontier Models Scored Under 1% for $8,900.

    Both cover AGI, ARC, Symbolica; reported by the same outlet (symbolica.ai); overlapping topics (architecture, model).

  7. Shared entities / Shared topic / What happened next

    Cursor Launches Composer 2 — Its Own Frontier Coding Model That Beats Claude Opus 4.6

    Both cover Anthropic, Google, OpenAI; overlapping topics (anthropic, beat, cost, model); picks up the Anthropic thread on 2026-03-20.

  8. Shared entities / Shared topic / What happened next / Tension

    Altman Pitches US-Led Forum to Set AI Standards

    Both cover Anthropic, Google, OpenAI; overlapping topics (anthropic, google); picks up the Anthropic thread on 2026-07-02.

Source trail

poetiq.aiSymbolica's parallel work

Entities

AGIAnthropicARCCruciallyDetailedGoogleLLMOpenAIPerson Team Beat GooglePoetiq ArchitectureSymbolica

Provenance

Canonical issue
Ramsay Research Agent — 2026-02-24
AI generated
no
Story unit
2026-02-24-poetiq-architecture-how-a-6-person-team-beat-google-on-arc-agi-2
Labels
source-backed, canonical briefing excerpt