BitNet 1.58B
by Microsoft Research · open-source · Last verified 2026-03-17
BitNet b1.58 is Microsoft Research's revolutionary 1-bit large language model where every weight is ternary {-1, 0, +1}, reducing memory and energy consumption dramatically while matching full-precision models in performance. This architecture represents a paradigm shift for efficient AI inference, enabling LLMs to run on CPUs and highly constrained edge devices.
https://huggingface.co/microsoft/bitnet_b1_58-3B ↗C
C—Below Average
Adoption: CQuality: B+Freshness: ACitations: BEngagement: F
Specifications
- License
- MIT
- Pricing
- open-source
- Capabilities
- text-generation, cpu-inference, ultra-low-memory-inference, instruction-following
- Integrations
- BitNet.cpp, Hugging Face
- Use Cases
- CPU-only inference, ultra-low-power edge devices, IoT AI deployment, energy-efficient AI serving
- API Available
- Yes
- Parameters
- ~3B
- Context Window
- 4K
- Modalities
- text
- Training Cutoff
- 2024
- Tags
- 1-bit, efficient, microsoft, quantization, edge
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
49.7Adoption
48
Quality
71
Freshness
80
Citations
65
Engagement
0