LFM2.5-24B-A2B

LFM2.5-24B-A2B is a Liquid Foundation Model from Liquid AI with 24 billion total parameters and about 2 billion active per token. It is positioned for fast on-device agents on high-end consumer hardware. Liquid AI announced it on 13 October (per its news feed, 2026). A YouTube review describes the weights as about 48 GB and reports that it beats similar-sized models such as Qwen 3.6 and Gemma 4 on average; this is a creator claim and not independently verified here.

The weights are open and released under Liquid's LFM Open License. According to the pricing page, the license is free for research, education, non-profit use, and for businesses with under $10M annual revenue. Companies above that threshold need a paid license. The license has no copyleft, so fine-tunes stay private. A reviewer described it as mostly Apache 2.0, but the official pricing page refers to the LFM Open License, so it is not an OSI-style open source license.

Liquid AI lists support for runtimes including llama.cpp, MLX, ONNX, Ollama, SGLang and vLLM. It also offers a hosted API at $0.05 per million input tokens (cached $0.025) and $0.70 per million output tokens. Enterprise customers can get bespoke models and on-prem deployment support. The exact modalities and context length were not given in the provided material; the model is treated here as text-in, text-out.