LiquidAI ships LFM2.5-24B-A2B: a 24B MoE with 2.45B active params for on-device agents
Skyline34rGt · reddit · 2026-10-14
LiquidAI released LFM2.5-24B-A2B, the final and largest model in its LFM2.5 family — a 24B-parameter MoE reasoning model with only 2.45B active parameters, built for fast, private on-device agents. GGUF weights are on Hugging Face.
- Agents: Competitive with Qwen3.6-35B-A3B (50% more total and active params) on agentic/tool-use tasks, scoring 48.09 vs 38.19 on LiveClaw.
- Speed: Thinks before answering with cheap reasoning tokens; up to 1.7x faster than comparable MoEs on-device — 184 tok/s on an Apple M5 Max, up to 21K output tok/s on a single NVIDIA B300.
- Specs: 40 layers (30 double-gated short convolution blocks + 10 GQA), 64 experts with top-4 routing, trained on 36T tokens, 131K context, 17 languages including Chinese.
- Suggested generation params: temperature 0.1, topk 50, repetitionpenalty 1.05.
Related event: LiquidAI Releases LFM2.5-24B-A2B On-Device MoE Model(3 posts)→
More from Models
- ChatGPT downweights self-promotional listicles, goes straight to Reddit for reviews — lilyraynyc · 2026-10-14
- First full Gemini 4 Argon hands-on review lands on YouTube: "Gemini is back" — Away_Cold_4943 · 2026-10-14
- Gemini 4 catches most fake API methods in hallucination stress test, one miss remains — haider1 · 2026-10-14
- Bindu Reddy predicts Fable 5.5 and Astra 6.1 launches within two weeks, with price cuts — bindureddy · 2026-10-14
- Pro members fear being left out as Gemini 4 'Argon' rumors swirl — MicahBerkley · 2026-10-14
- Google details safety evaluations and mitigations for Gemini 4 Argon launch — ancadianadragan · 2026-10-14