LiquidAI releases LFM2.5-24B-A2B reasoning model for on-device agents
LFM2.5-24B-A2B is an open-weight Mixture-of-Experts model that uses only 2B active parameters to provide fast, private reasoning on local hardware.
Summary
LiquidAI has introduced LFM2.5-24B-A2B, a 24B-parameter reasoning model designed for efficient on-device agentic tasks. Despite its 24B total parameters, it only activates 2B parameters per token, allowing for high-speed inference on consumer hardware like Apple M5 Max and NVIDIA GPUs. The model features an expanded 131K context window and is optimized for tool use and instruction following.
Technical specifications of LFM2.5-24B-A2B
- Architecture: 40 layers with 64 experts per MoE block using top-4 routing.
- Efficiency: Only 2B active parameters per token, enabling speeds up to 184 tok/s on Apple M5 Max.
- Context & Vocabulary: Context window increased to 131K tokens; vocabulary expanded to 128K tokens.
- Training: Pre-trained on ~33T tokens with specialized mid-training for reasoning, math, and code.
- Compatibility: Supports llama.cpp, MLX, vLLM, SGLang, and ONNX.
Why it matters
The model provides a high-performance, open-weight alternative for developers building private, local AI agents that require tool-calling capabilities without massive hardware requirements.
Original reference: huggingface.co →