Poetiq Inc., a startup working to make artificial intelligence large language models more capable and cost-efficient, has raised $45.8 million in seed funding.
FYRFLY Venture Partners and Surface Ventures jointly led the investment, which was announced on Thursday. Y Combinator, 468 Capital, Operator Collective, Hico Ventures and Neuron Venture Partners participated as well.
Poetiq was launched last year by former Google DeepMind researchers Shumeet Baluja and Ian Fischer. Baluja established the Alphabet unit’s computer vision group, while Fischer helped develop some of the technologies that underpin its LLMs. The company’s founding team also includes several other former researchers from the search giant.
Poetiq has developed a software platform that can enhance the capabilities of LLMs such as Google’s Gemini series and GPT-5.2. It also works with open-source models. According to the company, the platform uses a “meta-system” to identify ways of boosting LLMs’ output quality and reduce their inference costs.
Customers can apply the meta-system to a model by providing it with a few hundred training examples of the tasks they wish to automate. From there, Poetiq’s software turns the model into an artificial intelligence agent. The agent can automatically refine its own output quality, which is one of the means through which the platform boosts LLMs’ capabilities.
When a Poetiq-powered agent is given a task, it generates a potential answer and collects feedback. It then uses the underlying LLM to improve its answer based on the feedback. According to Poetiq, particularly complex requests prompt its platform to ask a sequence of questions that help it better understand the task at hand.
The company says its software can not only boost the quality of an LLM’s responses but also reduce the cost of generating them. It does so by determining when an LLM has collected all the information necessary to generate an answer and ending research queries, which avoids unnecessary calculations.
The company’s funding milestone comes a few weeks after it set a record on the ARC-AGI-2 benchmark. The benchmark, which is one of the toughest of its kind in the industry, comprises 1,000 visual puzzles designed to test LLMs’ reasoning capabilities. Poetiq says that its meta-system enabled GPT-5.2 to top the previous best score by 16%.
“For ARC-AGI 1 and 2, we used recursive self-improvement to produce specialized agents in a matter of hours,” Baluja said.
Startup AI21 Inc., which also offers a platform that makes existing LLMs more capable, is reportedly in talks with Nvidia Corp. about a potential acquisition. It’s believed a deal could value the software maker at up to $3 billion.
Image: Unsplash
Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.
- 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more
- 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network.
About News Media
Founded by tech visionaries John Furrier and Dave Vellante, News Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.
