Founded in June 2025 by former Google DeepMind researchers Shumeet Baluja and Ian Fischer, Poetiq builds a meta-system that pairs with any frontier LLM—such as GPT-5.2, Gemini, Claude, or Llama—to make the underlying model learn faster, reason more deeply, and operate at lower cost. Instead of requiring thousands of fine-tuning examples, customers supply only a few hundred samples, after which Poetiq’s system spins up an agent that recursively refines itself for the target problem. The technology recently set a new state-of-the-art on the ARC-AGI-2 benchmark, achieving 75% accuracy on the public evaluation set using GPT-5.2 and eclipsing prior leaders at roughly half the cost per task. An August 2025 MIT study highlighting that 95% of enterprises see no ROI on GenAI underscores the commercial need Poetiq addresses. The company positions itself as model-agnostic, augmenting rather than competing with frontier providers, which broadens its potential customer base across diverse AI platforms and use cases. Poetiq currently employs six people and is headquartered in Mountain View, California. To date, it has secured $45.8 million in seed financing to fuel product development and commercialization.

Total raised

$46M

Funding rounds

1

Latest round

Seed

Latest activity

Jan 2026