THE CRUNCH
Liquid AI has released LFM2.5-24B-A2B, an open-weight reasoning model designed to run agents on your own hardware. It has 24 billion total parameters but activates only 2 billion per token, which keeps each reasoning step cheap enough for continuous, private on-device use. The model thinks through an explicit chain of thought before answering, calls tools, and works through multi-step tasks inside popular agent harnesses.
It builds on LFM2-24B-A2B with the same architecture, including 40 layers and 64 experts per mixture-of-experts block with top-4 routing, but the context window grows from 32K to 131K tokens and the vocabulary doubles to 128K tokens for better non-Latin tokenisation. Pre-training scaled up from roughly 17 trillion to 33 trillion tokens, with tool calls and multi-turn trajectories mixed into training.
Liquid AI benchmarked it against Qwen3.6-35B-A3B, Gemma-4-26B-A4B-it and Nemotron-3.5-Lightning across 16 benchmarks. Despite being the smallest model in the group, it leads or matches the others on five of eight tool-use and agentic benchmarks and on all three instruction-following tests, scoring more than three times the next-best model on AutomationBench. It is also the fastest of the compared MoEs, decoding 184 tokens per second on an M5 Max and 114 on a Ryzen AI Max+ 395, up to 1.7 times faster than the next-best model, while using the least memory.
There are limits worth knowing: larger models keep an edge on knowledge-heavy benchmarks such as GPQA Diamond and on coding tests like SWE-Bench Verified, so the company suggests pairing it with retrieval or web search for those workloads. Day-one support covers llama.cpp, MLX, vLLM, SGLang and ONNX, and both the instruct and base versions are available on Hugging Face now.