LFM 2.5 Agentic Model for Tool Calling

Liquid AI released LFM 2.5, an 8-billion-parameter agentic model with 1 billion active parameters focused exclusively on tool calling, achieving nearly 200 tokens per second on a GPU. The model excels at routing users to appropriate MCP tools but is not suitable for complex text analysis or general-purpose tasks.

Large and powerful LLMs are great, but sometimes a smaller , local model is sufficient. For example, with an MCP server, a small local model can call tools and be used to accomplish things. I continued to search for small agentic models and found LFM 2.5 . This is a model created by Liquid AI , an American company, and it is focused on agentic tool calling . I downloaded the LFM2.5-8B-A1B-UD-Q4 K XL.gguf file and tested in llama-cpp . Basically this model is focused on tool calling exclusively. This means: • For complex analysis of text, LFM is not a good choice. • For trying to create a customer service AI system, where users indicate what their problem is, and the model routes them to the appropriate MCP tools, it is a good choice. I found that with a few more MCP server tools, I could get LFM 2.5 to find available tools and then call them based on my messages. The model has 8 billion parameters, but only 1 billion active, so it is extremely fast —I get nearly 200 tokens per second with a GPU, and it even works acceptably without a GPU. I have found LFM 2.5 to be an excellent agentic tool calling model, but beyond that, a larger model would be needed.