Poetiq emerged from stealth with measurable technical results and a clear strategic narrative, positioning its platform as an AI meta-system that changes how large language models reason, learn, and deliver enterprise value. At the same time, the startup announced it raised $45.8 million in Seed funding co-led by FYRFLY Venture Partners and Surface Ventures, with Y Combinator, 468 Capital, Operator Collective, Hico Ventures, and Neuron Venture Partners participating.
Notably, the funding announcement followed Poetiq’s performance on ARC-AGI-2, a benchmark designed to evaluate machine reasoning and progress toward artificial general intelligence. In early December, the company established a state-of-the-art result on the ARC-AGI-2 semi-private evaluation set, surpassing Gemini 3 Deep Think at half the cost per task using its system on top of Gemini 3 Pro. Shortly afterward, Poetiq integrated GPT-5.2 and reported a new state-of-the-art result of 75% accuracy on the public evaluation set, representing a 16 percentage point improvement over the previous benchmark.
Poetiq said its technology pairs with frontier models, including OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, and Meta’s Llama. Rather than requiring thousands or millions of examples for fine-tuning or reinforcement learning post-training, clients provide a problem and a few hundred examples. In response, the system generates a specialized agent and recursively improves it to increase accuracy and cost efficiency.
Consequently, the company framed its approach as a response to growing enterprise frustration with the cost and pace of model improvement. Moreover, it argued that accelerating reasoning performance without retraining models could fundamentally alter how organizations deploy and scale artificial intelligence.
Founders highlight structural limits in current model development
Poetiq was founded in June 2025 by Co-CEOs Shumeet Baluja, PhD, and Ian Fischer, former AI researchers at Google DeepMind.
During their collaboration at Google DeepMind, the founders identified persistent limitations in frontier models. Specifically, they observed that large language models struggled with many hard and easy problems, while reinforcement learning-based pre-training and post-training required weeks and significant financial resources.
“LLMs are impressive databases that encode a vast amount of humanity’s collective knowledge,” said Shumeet Baluja, co-CEO of Poetiq. “They are simply not the best tools for deep reasoning. That’s why efforts to improve their problem-solving skills are so slow and expensive. For ARC-AGI 1 and 2, we used recursive self-improvement to produce specialized agents in a matter of hours. It demonstrates how much we can help with problems that have been too hard or too expensive for LLMs alone.”
Poetiq positioned its technology as an alternative to the dominant strategy of scaling model size and training cycles. In addition, it argued that recursive specialization could accelerate reasoning performance while controlling operational costs.
Investors emphasize platform neutrality and enterprise relevance
Poetiq’s positioning appears distinctive in a crowded AI landscape. “That Poetiq managed to top ARC-AGI within six months of launching is remarkable,” said Philipp Stauffer, General Partner at FYRFLY Venture Partners. “Rather than compete against frontier models, their team of six found a way to coax more intelligence from every LLM available. Poetiq will be a must-have for companies trying to make AI work for real-world business applications.”
Gyan Kapur, co-Managing Partner at Surface Ventures, highlighted the company’s neutrality across platforms. “Poetiq is one of the rare AI startups that doesn’t need to outcompete frontier models or pick sides,” he said. “It can enhance any combination of LLMs, any native AI platform, and any AI use case. Poetiq can provide better performance at lower costs across diverse use cases by sitting on top of foundation models, and that is a unique position to be in.”
Benchmark progress signals strategic shifts in AI adoption
In this context, Poetiq’s benchmark results carried both symbolic and practical implications. OpenAI co-founder and President Greg Brockman commented publicly on the achievement, writing that Poetiq is “exceeding the human baseline on ARC-AGI-2 with gpt-5.2.”
Taken together, the announcement reflected a broader shift in AI strategy. As enterprises increasingly seek measurable returns on AI investments, solutions that enhance existing models rather than replace them could reshape procurement decisions and system architectures. Ultimately, Poetiq argued that the next phase of AI competition will focus less on model scale and more on orchestration, reasoning efficiency, and economic viability, signaling a potential realignment of priorities across the global AI ecosystem.
