Artificial Atlas

Who is betting on what in AI, and against whom.

Poetiq

Test-time reasoning wrapper; ARC-AGI-2 record.

HQ
Mountain View
Founded
Jun 2025
Web
poetiq.ai
Camps
Other substrates

Edit this entry

What it does

Two ex-DeepMind researchers built a 'meta-system' that wraps a frontier model in learned test-time reasoning. In December 2025 it set the verified ARC-AGI-2 record at 54%, on top of Gemini 3, at about half the cost of Google's own Deep Think. Raised a $45.8M seed in January 2026.

Camp membership over time

  1. dates not recorded
    Program search and other substrates
    Learned test-time reasoning over frontier models; ARC-AGI-2 record 54%.
    Sources Poetiq shatters ARC-AGI-2 state of the art at half the cost 2025-12-05 · Poetiq nabs $45.8M in seed funding for its LLM-enhancing ‘meta-system’ SiliconANGLE 2026-01-30

Capital

AmountKindValuationDateInvestorsSource
$46Mseed Jan 2026 FYRFLY Venture Partners, Surface Ventures, Y Combinator, 468 Capital, Operator Collective SiliconANGLE

People

SBShumeet BalujaCo-founder and co-CEO

General sources

Poetiq nabs $45.8M in seed funding for its LLM-enhancing ‘meta-system’ SiliconANGLE 2026-01-30