Liquid AI releases open-weight LFM2.5-24B-A2B for local agents

Liquid AI has released LFM2.5-24B-A2B, an open-weight, text-only reasoning model for local agents. The official article describes it as a 24B-total-parameter model with about 2B active parameters per token; the model card gives the precise active count as 2.45B. It is the largest and final release in the LFM2.5 family.

This is more than a new checkpoint. The model card lists a 131,072-token context window, function calling, and native, FP8, GGUF, ONNX, and MLX formats. Developers can select Transformers, vLLM, SGLang, llama.cpp, LM Studio, or Apple Silicon MLX based on their hardware and runtime rather than being restricted to a hosted API.

Adoption still needs workload-specific validation. The card recommends retrieval or web search for knowledge-heavy questions, and vendor benchmarks or throughput figures do not replace measurements under your quantization, memory, prompt length, and tool loop. The weights use Liquid AI's lfm1.0 license, which teams should review before redistribution or commercial deployment.

Start with a small typed tool set locally, record call validity, recovery from tool failures, and latency, then decide whether to connect the model to automations with real side effects.