This paper introduces SIMPL (Simple and efficient Motion Prediction baseLine), a novel multi-agent motion prediction method for autonomous driving. SIMPL addresses the challenges of context encoding, symmetric scene modeling, and trajectory representation. It employs an instance-centric scene representation and a symmetric fusion Transformer (SFT) to efficiently fuse global features, enabling real-time and accurate predictions for all road users. The method uses Bernstein basis polynomials for continuous trajectory parameterization, ensuring smoothness and facilitating the evaluation of higher-order derivatives. SIMPL is evaluated on the Argoverse 1 & 2 datasets, demonstrating competitive performance compared to state-of-the-art methods while offering a lightweight design and low inference latency. The code is open-sourced, making it highly extensible and promising for real-world deployment.This paper introduces SIMPL (Simple and efficient Motion Prediction baseLine), a novel multi-agent motion prediction method for autonomous driving. SIMPL addresses the challenges of context encoding, symmetric scene modeling, and trajectory representation. It employs an instance-centric scene representation and a symmetric fusion Transformer (SFT) to efficiently fuse global features, enabling real-time and accurate predictions for all road users. The method uses Bernstein basis polynomials for continuous trajectory parameterization, ensuring smoothness and facilitating the evaluation of higher-order derivatives. SIMPL is evaluated on the Argoverse 1 & 2 datasets, demonstrating competitive performance compared to state-of-the-art methods while offering a lightweight design and low inference latency. The code is open-sourced, making it highly extensible and promising for real-world deployment.