According to DeepLearning.AI, Samsung’s Tiny Recursive Model (TRM) iteratively refines answers with a running context of past changes to solve structured grid puzzles such as Sudoku, Mazes, and ARC-AGI tasks (source: DeepLearning.AI on X, Dec 17, 2025). According to DeepLearning.AI, TRM tops many LLMs, including DeepSeek-R1 and Gemini 2.5 Pro, on these benchmarks, highlighting competitive gains in reasoning performance relevant to AI-focused traders tracking benchmark leadership (source: DeepLearning.AI on X, Dec 17, 2025).