Nanbeige Releases 3B Chinese-Enhanced Language Model
The new Llama-based model was trained from scratch on 3.5 trillion tokens of Chinese and English data to enhance its bilingual capabilities.
Chinese AI firm Nanbeige Technology has released Nanbeige4.1-3B, a new 3-billion-parameter language model designed for strong performance in both Chinese and English. The model is based on the popular Llama architecture but was trained from scratch on a custom, high-quality dataset of 3.5 trillion tokens.
This release adds a new contender to the growing field of specialized, efficient open models. By focusing on a specific language pair and a compact size, Nanbeige4.1-3B offers developers a capable option that can run on less powerful hardware. The company's provided benchmarks show it performing competitively against other models in its class on Chinese-centric evaluations like CMMLU and C-Eval.
Key Specifications
The model's technical details make it a practical choice for a range of applications:
- Architecture: Llama
- Parameters: 3 billion
- Context Length: 4096 tokens
- Training Data: 3.5T tokens (Chinese & English)
According to its Hugging Face release card, Nanbeige4.1-3B is available under a custom license that permits free commercial use, an important consideration for teams looking to build products with the model.
Sources
- Visit
Nanbeige/Nanbeige4.1-3B
Hugging Face
More in Text / LLM
Meituan Ships a Lighter, Sparser LongCat-Flash
The food-delivery giant's newest open model trims its mixture-of-experts design for more efficient inference under an MIT license.
DeepSeek Refreshes V4-Flash With New 0731 Checkpoint
The MIT-licensed mixture-of-experts model returns in an updated build shipping with FP8 weights for cheaper inference.
DeepSeek Ships V4-Flash, a 304B MoE Tuned for Agents
The latest checkpoint in DeepSeek's V4 line leans into agentic workflows while keeping the permissive MIT license.
0 comments
No comments yet. Be the first to weigh in.