Hugging Face has released LFM3, a new family of Liquid Foundation Models designed for specialized intelligence. These models feature a tri-hybrid architecture and a new training recipe, enabling LFM3-1.7B to outperform LFM2.5-2.6B with fewer parameters. LFM3 supports multiple modalities including text, vision, and audio on a single backbone and is optimized for edge deployment, with the text model fitting into approximately 1 GB at 4-bit precision and a 256K context window for on-device agents. The models are built for specialization, allowing fine-tuning on a single GPU for specific data and terminology. AI
IMPACT Enables specialized multimodal AI on edge devices with improved parameter efficiency.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- LFM2
- LFM2-1.2B
- LFM2.5-2.6B
- LFM2.5-8B-A1B
- LFM3
- LFM3-1.7B
- LFM3-1.7B-Base
- Liquid Foundation Models
- Qualcomm Hexagon NPU
- Qwen3.5-2B
- Raspberry Pi 5
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
