Souper-Model: How Simple Arithmetic Unlocks State-of-the-Art LLM Performance - Explained Simply | ArXiv Explained