Liquid AI Releases LFM2.5-24B-A2B Reasoning Model
Liquid AI introduces LFM2.5-24B-A2B, a 24B Mixture-of-Experts reasoning model with 2B active parameters tailored for local agents.
Stock photo for illustration only, not from the actual event
- Liquid AI releases open-weight LFM2.5-24B-A2B model
- Features a 24B Mixture-of-Experts architecture with 2B active parameters
- Targeted specifically for local on-device AI agents
- Achieves 184 tokens per second decoding speed on Apple M5 Max
Liquid AI has officially launched LFM2.5-24B-A2B, an open-weight Mixture-of-Experts (MoE) reasoning model engineered to empower local agents operating directly on user hardware. The model focuses on delivering high-performance reasoning capabilities while maintaining efficiency for local deployment.
The architecture consists of 24 billion total parameters while utilizing only 2 billion active parameters during inference. This selective activation mechanism allows the model to maintain the expressive capacity of a large model while operating with the speed and efficiency closer to smaller models.
Stock photo for illustration only, not from the actual event
In benchmark evaluations, the LFM2.5-24B-A2B model scored 48.09 on LiveClaw. Furthermore, benchmark tests running on an Apple M5 Max processor demonstrated a decoding performance of 184 tokens per second, highlighting its capability for demanding local workloads.
Sparse Mixture-of-Experts architectures represent a crucial shift in local AI deployment, allowing developers to pack extensive knowledge into a large parameter pool while activating only a fraction of those parameters per token to preserve computational efficiency and speed on consumer hardware.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment