Liquid AI's LFM2.5-24B-A2B is a 24B-parameter reasoning model with 2B active parameters, built for fast, private agents on-device. It supports tool calls and multi-step tasks.
LFM2.5-24B-A2B is a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device.
The context window grows from 32K to 131K tokens, and an in-place tokenizer expansion doubles the vocabulary from 65K to 128K tokens.
LFM2.5-24B-A2B is the fastest of the compared MoEs, decoding 184 tokens/s on an M5 Max and 114 tokens/s on a Ryzen AI Max+ 395.
Both LFM2.5-24B-A2B and LFM2.5-24B-A2B-Base are open-weight and available on Hugging Face today.
Hugging Face published a blog post from Liquid AI introducing LFM2.5-24B-A2B, a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device. The model builds on LFM2-24B-A2B, keeping its architecture: 40 layers, 64 experts per MoE block with top-4 routing, and the same hybrid of gated short convolutions and GQA. The context window grows from 32K to 131K tokens, and an in-place tokenizer expansion doubles the vocabulary from 65K to 128K tokens for better non-Latin tokenization. Unlike its predecessor, it is a reasoning-only model that produces an explicit chain of thought before its final answer. Pre-training was scaled up from ~17T to ~33T tokens, with mid-training adding 2T tokens at 32K context and 300B tokens at 131K context.
Because only 2B parameters are active per token, each reasoning token is cheap, so you can run agents continuously on your own hardware. The model is designed to think before it answers, call tools, and work through multi-step tasks inside popular agent harnesses. It ships with day-one support across the inference ecosystem, including llama.cpp, MLX, vLLM, SGLang, and ONNX. Edge inference results show it decoding 184 tokens/s on an M5 Max and 114 tokens/s on a Ryzen AI Max+ 395, up to 1.7x faster than the next-best model while using the least memory. GPU inference reaches over 21K output tokens per second at high concurrency on a single B300. Both LFM2.5-24B-A2B and LFM2.5-24B-A2B-Base are open-weight and available on Hugging Face today.
The blog post reports that LFM2.5-24B-A2B was evaluated against Qwen3.6-35B-A3B, Gemma-4-26B-A4B-it, and Nemotron-3.5-Lightning on 16 benchmarks covering tool use and agentic workflows, knowledge, instruction following, math, and coding. It is the smallest model in the group, yet competitive with MoEs with ~50% more total and active parameters. The model leads or matches the competition on five of the eight tool-use and agentic benchmarks (AutomationBench, BFCLv4, ToolSandbox, LiveClaw, and τ³-Bench Banking), and on AutomationBench it scores more than three times the next-best model. It also leads on all three instruction-following benchmarks. Larger models keep an edge on knowledge-heavy benchmarks like GPQA Diamond, HLE, and AA-Omniscience, and on coding benchmarks like SWE-Bench Verified (OpenCode) and SciCode.
The supplied feed entry is a concise summary; consult the canonical article for full context. The blog post mentions post-training turns the base model into an agent in five stages, but the specific details of those stages are not elaborated in the summary. The post also references a release blog with video examples and a guide for full setup, but those resources are not included in the summary. The exact benchmark scores for each model comparison are not fully detailed in the summary, only the relative outcomes. The compatibility requirement with transformers>=5.0.0 is stated, but the specific installation and usage code examples are not included in the summary text.
LFM2.5-24B-A2B is a 24B-parameter reasoning model with just 2B active parameters, built for fast, private agents on-device. It thinks before it answers, calls tools, and works through multi-step tasks inside popular agent harnesses. Because only 2B parameters are active per token, each reasoning token is cheap, so you can run agents continuously on your own hardware.
The blog post reports that LFM2.5-24B-A2B was evaluated against Qwen3.6-35B-A3B, Gemma-4-26B-A4B-it, and Nemotron-3.5-Lightning on 16 benchmarks. It is the smallest model in the group, yet competitive with MoEs with ~50% more total and active parameters. It leads or matches the competition on five of the eight tool-use and agentic benchmarks, and leads on all three instruction-following benchmarks.