Poetiq Secures $45.8M to Enhance AI Technologies
Founded by former Google DeepMind scientists, Poetiq emerges from "stealth" after setting a new state-of-the-art (SOTA) on ARC-AGI-2.
Poetiq, the innovative developer of an AI meta-system designed to boost the performance of large language models (LLMs), has recently announced an impressive funding round of $45.8 million in Seed funding. This announcement highlights the company’s potential to push the boundaries of artificial intelligence and problem-solving. The funding was co-led by prominent investors such as FYRFLY Venture Partners and Surface Ventures, while Y Combinator, 468 Capital, Operator Collective, Hico Ventures, and Neuron Venture Partners also contributed.
Setting New Standards in AI Performance
The funding comes on the heels of Poetiq achieving remarkable results on the ARC-AGI-2 benchmark, which evaluates the potential of AI systems in reasoning and approaching artificial general intelligence (AGI). Poetiq’s ability to pair seamlessly with frontier LLMs, including notable models like OpenAI's Chat GPT, Anthropic's Claude, and Google's Gemini, allows it to address complex challenges more effectively, needing far fewer examples to learn.
In practice, clients share a specific problem with Poetiq, alongside a few hundred examples, as opposed to the traditional requirement of thousands or millions of data points for fine-tuning. This innovative approach leads to the generation of specialized agents that not only address the immediate issues but also enhance their performance incrementally through recursive improvements.
Founders' Vision and Experience
Founded in June 2025 by co-CEOs Shumeet Baluja, PhD, and Ian Fischer, Poetiq leverages a wealth of experience from the AI research community. Baluja, a former CTO at Jamdat Mobile with over 21 years at Google DeepMind, holds more than 170 patents related to neural networks and machine learning. Fischer, who joined Google DeepMind following its acquisition of Apportable, brings a strong background in AI development.
Together, they recognized challenges faced by frontier LLMs, which often faltered in addressing both hard and simple problems effectively. The existing methodologies—primarily reliant on extensive pre-training and reinforcement learning—were often costly and time-prohibitive, making them accessible primarily to large corporations.
Transforming AI Solutions
"LLMs are impressive databases that encode a vast amount of humanity's collective knowledge," stated Shumeet Baluja. "However, they are not necessarily the optimal tools for advanced reasoning. This is why enhancing their problem-solving capabilities has been slow and expensive. Our innovative recursive self-improvement approach allows specialized agents to be produced in just hours, demonstrating how easily we can tackle challenges that have previously been deemed too complex or costly for LLMs alone."
Recent studies underscore the importance of Poetiq's approach. An MIT study from 2025 revealed that despite substantial investments of $30 to $40 billion in Generative AI, a staggering 95% of organizations reported receiving zero return on their investment. This gap highlights the demand for solutions like Poetiq, which can notably improve reasoning capabilities across various machine learning models.
A Unique Competitive Advantage
Philipp Stauffer, General Partner at FYRFLY Venture Partners, remarked on the significance of Poetiq's rapid ascendancy. "The fact that Poetiq managed to outperform the ARC-AGI benchmark within just six months of its launch is absolutely remarkable. By focusing on extracting greater intelligence from existing LLMs rather than competing directly, Poetiq has positioned itself as an essential tool for businesses aiming to integrate AI into practical applications."
Similarly, Gyan Kapur, co-Managing Partner at Surface Ventures, emphasized Poetiq's versatility. "What sets Poetiq apart is its ability to enhance a combination of LLMs and any native AI platforms, avoiding the need to side with specific models. This strategic advantage enables better performance across various AI applications—ultimately at a lower cost."
In the last quarter of 2025, Poetiq notably set a new standard on the ARC-AGI-2 evaluations, surpassing the previous leader, Gemini 3 Deep Think, at only half the cost per task. The swift incorporation of OpenAI's latest model, GPT-5.2, into their system further exemplifies Poetiq's agility, attaining a 75% accuracy rate on the public evaluation set, signaling a significant advancement in AI capabilities.
As Poetiq continues to garner attention for its trailblazing innovations in AI technology, the industry eagerly anticipates its next steps and the potential impact on both the tech landscape and business applications.
Frequently Asked Questions
What is Poetiq's main innovation?
Poetiq's main innovation lies in its AI meta-system that enhances the performance of LLMs, allowing them to learn quicker and address complex problems efficiently.
Who are the founders of Poetiq?
Poetiq was founded by Shumeet Baluja and Ian Fischer, both of whom have extensive experience in AI research, particularly at Google DeepMind.
How does Poetiq differ from traditional LLMs?
Unlike traditional LLMs that require massive datasets for training, Poetiq needs just a few hundred examples, generating specialized agents that learn more efficiently through recursive improvements.
What benchmarks has Poetiq achieved?
Poetiq recently established a new state-of-the-art on the ARC-AGI-2 benchmark, outpacing competitors like Gemini 3 Deep Think and achieving high accuracy with lower costs.
Why do businesses need solutions like Poetiq?
Many businesses find it difficult to achieve a return on investment from existing AI technologies. Poetiq offers a way to enhance reasoning capabilities and improve overall performance without significant additional investments.
