# Leading in multi-domain benchmarks

A quantized version model developed based on DeepSeek-R1-0528-Qwen3-8B, with significant improvements in reasoning ability and reduction of hallucination rate, suitable for various natural language processing tasks.

A 7B-parameter large language model based on the Mistral architecture, trained with DPO (Direct Preference Optimization), demonstrating excellent performance in multiple benchmarks