According to CoinMeta, the Italian AI laboratory Paradigma has released its first model limite with 1 billion parameters, specifically designed for solving math problems. The model was trained from scratch and used less than 300 billion tokens. After training through SFT and reinforcement learning, official evaluations show that this model scored 94.01% in AIME 2026, surpassing many models with 3B to 120B parameters. However, the scores of some competitors were directly taken from model cards or MathArena, and not from a unified retest by Paradigma; therefore, the evaluation configurations may have differed. Paradigma positions this model as a high-throughput math solver, mainly handling single-round math problems, and plans to use it as a high-frequency math solver within multi-agent systems. The model's weights and VLLM inference code have been made available according to the Apache 2.0 protocol.