VLANeXt is a research-oriented codebase that systematically explores design choices for vision-language-action models across more than 500 experiments spanning foundational components, perception essentials, and action modeling. The recipe combines a dedicated policy module with action chunking, flow matching, multi-view inputs, and frequency-domain regularization to build an efficient VLA system. Beyond the core recipe, the codebase includes six additional baseline variants covering smaller and larger backbones, latent action learning, latent-space prediction, and world-action modeling, giving researchers clean starting points for exploring robotics foundation models at different scales and with different learning objectives.