DeepSeek-V3 is a state-of-the-art open-source Mixture-of-Experts (MoE) large language model from DeepSeek AI, featuring 671B total parameters with 37B activated per token for highly efficient inference. The model achieves frontier-level benchmark performance — including an AIME 2025 score of 89.3 and a Chatbot Arena rating of 1421 — while being fully open-weight under the MIT license for commercial use. It represents a landmark achievement in closing the gap between open-source and proprietary frontier LLMs.