DeepSeek-V3, ultra-large open-source AI, outperforms Llama and Qwen on launch

Chinese AI startup DeepSeek has launched DeepSeek-V3, an ultra-large AI model boasting 671 billion parameters. Released on Hugging Face under DeepSeek’s license, this model leverages a mixture-of-experts (MoE) architecture, selectively activating parameters to improve efficiency and accuracy.

DeepSeek-V3 has already outperformed leading open-source models like Meta’s Llama 3.1-405B and shows performance on par with proprietary models from OpenAI and Anthropic. This advancement underscores the narrowing gap between open-source and closed AI models, with DeepSeek aiming to drive progress toward artificial general intelligence (AGI).