DeepSeek V3

DeepSeek V3 is the late-2024 open-weight MoE model from Chinese lab DeepSeek. Its documented training cost — surprisingly low for the performance level — sparked a wide industry debate about price-performance versus closed frontier models. Along with Llama 4, Qwen 3, and Mistral it sits at the center of the Open-source Race dynamic. The technical foundation behind the same lab's reasoning-focused DeepSeek R1 also traces back to the V3 line.