Boole-1.7B (precisionnumerics.ai)

The AI community sees the launch of Boole-1.7B, a pioneering 1-bit language model built from Qwen3-1.7B. Released on October 11, 2026, Boole-1.7B boasts a compact weight of 247 MB while achieving a score of 50.55 across ten benchmarks—85.9% of the BF16 baseline score of 58.83, representing significant performance efficiency with a remarkable 13.9× reduction in stored weight size. This model signifies a major step in the development of low-precision language models, addressing the challenges of binary weight representation. By simplifying weights to just signs (+1 or -1) without retaining magnitude, Boole seeks to maximize performance despite potential accuracy trade-offs associated with losing detailed weight information. Its architecture is designed for English inference tasks, and it notably outperforms previous binary models in multiple benchmarks, particularly in math and coding tasks. The introduction of Boole-1.7B could inspire further innovations in efficient model design while demonstrating the trade-offs between precision and resource usage in AI applications.