Run fast, private agents on-device with LFM2.5-24B-A2B
Best-in-class agent: Competitive with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks, scoring 48.09 on LiveClaw versus 38.19.
Proof1 community thread
Key points
- Reasoning at MoE speed: Always thinks before it answers.
- LFM2.5-24B-A2B builds on LFM2-24B-A2B and keeps its architecture: 40 layers, 64 experts per MoE block with top-4 routing, and the same hybrid of gated short convolutions and GQA.
- We evaluated LFM2.5-24B-A2B 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.
- "LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.
Sources (2)
- [1]Run fast, private agents on-device with LFM2.5-24B-A2BHugging Face Blog · Oct 13, 05:05 PM
- Best-in-class agent: Competitive with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks, scoring 48.09 on LiveClaw versus 38.19.
- Reasoning at MoE speed: Always thinks before it answers.
- [2]LiquidAI/LFM2.5-24B-A2Br/LocalLLaMA (top, daily) · Oct 13, 05:14 PM
"LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026unslothai/unsloth v0.1.907-beta: Large Performance Gains + More Browser Use
- Oct 13, 20264x 6000 Pro for the company
- Oct 12, 2026openbmb/MiniCPM-V-4.7-35B-A3B
- Oct 12, 2026unslothai/unsloth v0.1.906-beta: Browser Use + Qwen-Image-2.1-Turbo
- Oct 12, 2026LiquidAI/LFM2.5-24B-A2B
- Oct 9, 2026Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash