DeepSeek V3 is a highly performant and efficient Large Language Model (LLM) that has generated significant hype within the AI community due to its performance and operational cost combination. Its innovative features, including Multi-Head Latent Attention (MLA), Mixture of Experts (MoE), and Multi-Token Predictions (MTP), contribute to both efficiency and accuracy during training and inference phases. MLA compresses input embedding dimension into low-rank representation, saving KV cache memory and speeding up token generation. MoE speeds up the token generation process by activating only certain experts during inference, depending on the task. MTP can be repurposed for speculative decoding, enabling faster generation processes. DeepSeek V3's open-source nature under the MIT license enables the global AI community to contribute, experiment, and build upon its technology, accelerating progress toward Artificial General Intelligence (AGI). Its performance is already superior compared to other state-of-the-art LLMs, but research suggests that it can be further optimized with knowledge distillation techniques.