Tiny Recursive Model

Samsung’s researchers introduce the Tiny Recursive Model (TRM), a 7-million parameter AI model, challenging the “bigger is better” paradigm. TRM achieves state-of-the-art results on complex reasoning tasks, including ARC-AGI, surpassing larger models like Gemini 2.5 Pro. TRM’s single, compact network recursively refines reasoning and answers, achieving 87.4% accuracy on Sudoku-Extreme and 85.3% on Maze-Hard. This efficient model, building on the Hierarchical Reasoning Model, suggests a future for AI focused on smarter architectures rather than sheer scale.