Nanbeige LLM Lab has released Nanbeige4.1-3B, an open-source language model with 3 billion parameters. The primary goal of this model is to combine advanced reasoning capabilities, strong alignment with human preferences, and agentic functionalities, all within a small-sized model.

Key Features

  • Strong Reasoning: Nanbeige4.1-3B is designed to solve complex problems through sustained and coherent reasoning, achieving significant results in challenging tasks such as LiveCodeBench-Pro, IMO-Answer-Bench, and AIME 2026 I.
  • Alignment with Human Preferences: In addition to problem-solving, the model demonstrates strong alignment with human preferences, achieving high scores on Arena-Hard-v2 and Multi-Challenge.
  • Agentic Capabilities: Nanbeige4.1-3B natively supports agentic functionalities, including deep-search capabilities, with good performance on xBench-DeepSearch and GAIA.
  • Extended Context: The model supports contexts up to 256,000 tokens, allowing the management of complex tasks that require in-depth analysis and the use of numerous tools.

The model is available on Hugging Face. For those evaluating on-premise deployments, there are trade-offs to consider; AI-RADAR offers analytical frameworks on /llm-onpremise for evaluation.