One model DeepSeek-LLM, a 67B parameter language model, outperforms LLaMA2 70B in key tasks. Available for research and commercial use.