Elorian Launches Glimpse Visual Reasoning Model with Notable Results

This month, Elorian launched its first visual reasoning model, Glimpse, promoted as a new category of visual thinking model. It is designed to reason about the physical world based on visual input. Elorian partnered with AI data company Mercor to introduce the DecoyCount benchmark, testing whether language models accurately count objects or rely on visual shortcuts. Results showed frontier models scored below 50% accuracy, with GPT-6 Astra achieving the highest at 43.17%.