Liquid AI Releases LFM2.5-24B-A2B, an Efficient 24B Model for Laptops and PCs, with Open Weights

On October 14, Liquid AI announced the release of LFM2.5-24B-A2B, the largest member of the LFM2.5 family, and made its base and post-trained weights available on Hugging Face. Designed as an efficient local model for laptops and PCs, it has 24B total parameters with 2B active per token, a context window expanded to 131K tokens, a 128K-token vocabulary, and 33T pretraining tokens.

According to the company, the model is a reasoning model that thinks before answering, making it suitable for multi-step tasks in popular agent harnesses, and it is easy to fine-tune and customize. On agentic and tool-use tasks, it performs competitively with Qwen3.6-35B-A3B, a model with roughly 50% more total and active parameters. LFM2.5-24B-A2B is also now available on OpenRouter for developers.