Liquid AI Releases LFM2.5-24B-A2B for Local AI Agents

Liquid AI releases LFM2.5-24B-A2B, an open weight reasoning model for local agents, as detailed in a company blog post. The model has 24 billion total parameters, activates 2 billion parameters per token and supports a context window of 131,072 tokens.

Liquid AI says the model is competitive with Qwen3.6-35B-A3B, which has about 50% more total and active parameters, on five of eight agentic and tool use benchmarks. It scored 48.09 on LiveClaw and 27.69 on AutomationBench compared with 38.19 and 8.72 for the Qwen model.

The company measured generation speeds of 184 tokens per second on an Apple M5 Max and 114 tokens per second on an AMD Ryzen AI Max+ 395. On a single NVIDIA B300 GPU, it reached more than 21,000 output tokens per second under high concurrency.

The base and post trained versions are available through Hugging Face and Liquid AI's Playground. Supported inference software includes llama.cpp, MLX, vLLM, SGLang and ONNX.