Elorian’s Glimpse Model Beats Frontier AI on DecoyCount Visual‑Reasoning Benchmark

Elorian released its first visual‑reasoning model, Glimpse, in April after raising $55 million in seed funding. The company introduced the DecoyCount benchmark, which tests whether models can count objects accurately without relying on visual shortcuts. In a single run on 1,660 DecoyCount items, Glimpse achieved 50.48% accuracy, the only model in the comparison to exceed 50%. By contrast, the top frontier models scored below 43% – GPT‑6 Astra at 43.17%, Gemini‑3.8‑flash at 32.71%, Qwen3.8‑max at 28.31% and Claude‑Fable‑5.1 at 25.22%. Elorian confirmed that the DecoyCount images and prompts were not part of Glimpse’s training data, and that no contamination was detected. The benchmark highlights failures that separate vision and reasoning, which many existing tests conflate.