6thWave: AI News Hub

Chinese AI startup DeepSeek has unveiled DeepSeek Coder V2, an open-source mixture of experts (MoE) code language model that promises to outperform state-of-the-art closed models like GPT-4 Turbo, Claude 3 Opus, and Gemini 1.5 Pro. Built on the foundation of the DeepSeek-V2 model and trained on a massive dataset of 6 trillion tokens, DeepSeek Coder V2 supports over 300 programming languages, making significant strides in coding and math tasks. It features a context window of 128K, allowing it to handle more complex coding challenges. The model has shown superior performance across various benchmarks, establishing itself as a leader in the open-source domain. DeepSeek’s approach uses a Mixture of Experts framework to activate specific parameters for optimized performance. Additionally, the model maintains strong general language and reasoning capabilities. DeepSeek Coder V2 is available under an MIT license, promoting both research and commercial use.

Source.