A new open source model that delivers leading performance and quality across text-based use cases such as synthetic data generation at a fraction of the inference cost.