Deepseek V3

Powerful 671B parameter MoE language model with state-of-the-art performance.

DeepSeek v3 is a powerful 671B parameter Mixture-of-Experts (MoE) language model that offers groundbreaking performance. It is an AI-driven LLM with 671B total parameters (37B activated per token) and supports API access, an online demo, and research papers. Pre-trained on 14.8 trillion high-quality tokens, DeepSeek v3 delivers state-of-the-art results across various benchmarks, including mathematics, coding, and multilingual tasks, while maintaining efficient inference. It features a 128K context window and incorporates Multi-Token Prediction for enhanced performance and acceleration.