Poetiq
Category: AI Infrastructure
An AI meta-system designed to enhance the reasoning and problem-solving capabilities of existing large language models (LLMs) through recursive self-improvement. Poetiq was founded in 2025. The company is led by Shumeet Baluja & Ian Fischer. Based in Mountain View, CA. Team size: 6-10. Total funding raised: $45.8M. Latest round: Seed. Key investors include FYRFLY Venture Partners, Surface Ventures, Y Combinator, 468 Capital, Operator Collective.
- Founded
- 2025
- Headquarters
- Mountain View, CA
- Team size
- 6-10
- Total funding
- $45.8M
Value proposition
Significantly boosts the reasoning performance of frontier models like GPT-4 and Gemini without the prohibitive costs and time required for model retraining.
Products and solutions
Poetiq Reasoning Meta-System, Recursive Self-Improvement Framework, Augur (Model Routing / Benchmark / Training Data), ARC-AGI-2 Optimization Engine, Cross-Model Reasoning Enhancement Layer
Unique value
Unlike traditional AI startups that train new base models, Poetiq builds a 'meta-layer' that enables existing models to solve complex, novel problems they were not specifically trained for.
Target customer
Enterprise AI research labs, software engineering firms, and organizations requiring advanced logical reasoning beyond standard LLM capabilities.
Industries served
Artificial Intelligence, Software Development, Scientific Research, Data Science & Analytics
Technology advantage
Recursive self-improvement meta-system that lifts frontier LLMs without retraining: ARC-AGI-2 54%+ verified / 75% with GPT-5.2; SOTA on HLE, SimpleQA, LiveCodeBench Pro; Augur maps model failure boundaries for routing and training data.
How they differentiate
Unlike competitors that train massive base models from scratch, Poetiq develops a 'meta-system' that layers onto existing frontier models (like Gemini or GPT-4). It uses recursive self-improvement to enhance reasoning without the need for expensive retraining or massive compute overhead.
Main competitors
OpenAI (o1/Strawberry), Sakana AI, Anthropic, Cognition AI
Key partnerships
Y Combinator (W25 Batch), FYRFLY Venture Partners, Surface Ventures, 468 Capital, Operator Collective
Major milestones
Raised $45.8M Seed round (Jan 29, 2026), ARC-AGI-2 SOTA: 54%+ verified semi-private; 75% public eval with GPT-5.2, SOTA on Humanity's Last Exam and SimpleQA (Feb 2026), LiveCodeBench Pro coding SOTA via Meta-System (May 2026), Launched Augur product for model routing and benchmark creation (Aug 2026), Y Combinator backed; founding team ex-Google DeepMind senior researchers
Growth metrics
ARC-AGI-2: 54%+ verified semi-private SOTA and 75% public eval with GPT-5.2 (exceeding ~60% human baseline); subsequent SOTAs on HLE, SimpleQA, and LiveCodeBench Pro; Augur product launched Aug 2026.
Market positioning
Advanced AI reasoning infrastructure provider focusing on 'System 2' thinking capabilities.
Geographic focus
United States (North America)
Patents and IP
No specific registered patents disclosed; proprietary algorithms focus on recursive logic and meta-reasoning architectures.
About Shumeet Baluja & Ian Fischer
Shumeet Baluja was a Senior Staff Research Scientist at Google DeepMind and Google Research for over 20 years, holding a PhD from CMU. Ian Fischer was a Senior Staff Research Scientist at Google DeepMind and an OpenAI researcher, previously co-founding Apportable (acquired by Google).
Latest news about Poetiq
More AI Infrastructure companies
Official website: https://poetiq.ai/