Microsoft Researchers Develop Hyper-Efficient AI Model That Can Run On CPUs

Microsoft has introduced BitNet b1.58 2B4T, a large-scale 1-bit AI model with 2 billion parameters that can run efficiently on CPUs. The model is openly available under an MIT license, allowing for widespread use and development. According to Microsoft researchers, BitNet b1.58 2B4T is the first bitnet with 2 billion parameters, which are largely synonymous with weights. The model was trained on a massive dataset of 4 trillion tokens, equivalent to about 33 million books. This training enables BitNet b1.58 2B4T to outperform traditional models of similar sizes, as claimed by the researchers. The model's performance is notable, as it holds its own against rival 2 billion-parameter models, although it does not significantly outperform them. BitNet b1.58 2B4T surpasses other models, including Meta's Llama 3.2 1B and Google's Gemma 3 1B, on certain benchmarks. The model's speed and efficiency are also impressive, as it is twice as fast as other models of its size while using a fraction of the memory. However, achieving this performance requires using Microsoft's custom framework, bitnet.cpp, which currently only works with certain hardware. The framework's limited compatibility, excluding GPUs, may be a significant limitation for widespread adoption of BitNet b1.58 2B4T.