LFM2.5-24B-A2B on-device agent model released
The gist
Liquid AI released LFM2.5-24B-A2B, a 24B-parameter model activating 2B per token.
It targets private on-device agents, with open weights and day-one ecosystem support.
How it unfolded
- Oct 13, 2026Liquid AI released LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters, built for fast, private agents on-device; the base and post-trained models are open-weight on Hugging Face and the company's Playground.
- Oct 14, 2026MarkTechPost described it as a 2B-active MoE for on-device agents, and Startup Fortune reported the October 13 ship, noting it activates only 2 billion parameters per token.
- Oct 15, 2026LavX News and Tech Insider Australia reported the release, including 184 tokens/s on an Apple M5 Max, a 131,072-token context window, and day-one support for llama.cpp (GGUF), MLX, vLLM, SGLang, and ONNX.