1 article found in this topic.
Boss Zhipin's Nanbeige LLM Lab has unveiled Nanbeige4-3B, a 3-billion-parameter small language model (SLM) designed to offer faster inference and lower deployment costs than larger LLMs. It reportedly outperforms models like Qwen3-4B and Qwen3-8B in various benchmarks, demonstrating competitive capabilities even against trillion-parameter models in creative tasks. The model's development involved extensive pre-training with 23 trillion tokens and a multi-stage post-training process.
