Most powerful 7B large language model. Supports English and code with 8K context. Matches GPT-4 performance on some benchmarks. Built on top of Mistral 7B, enhanced with additional pre-training and fine-tuning for strong problem-solving capabilities.