The Llamba models are part of Cartesia's Edge library, designed for efficient, high-performance machine learning applications. These recurrent models leverage distillation techniques to achieve strong performance while maintaining computational efficiency across various scales, including 1B, 3B, and 8B parameter variants. The Llamba models have been evaluated on standard benchmarks including ARC, PIQA, Winogrande, HellaSwag, Lambada, MMLU, and OpenBookQA, demonstrating their capability to maintain competitive performance across multiple language understanding tasks.