Liquid AI 发布 LFM2.5-24B-A2B，已上架 OpenRouter
Liquid AI 发了个能在笔记本上本地跑的推理模型，24B 参数只激活 2B，做智能体任务能打比它大一半的 Qwen3.6-35B-A3B，权重全开源。
Liquid AI 发布 LFM2.5 家族最大成员 LFM2.5-24B-A2B，现已在 OpenRouter 上架。该模型为 MoE 架构，总参数 24B、每 token 激活 2B，上下文扩展至 131K tokens，词表 128K，预训练数据 33T tokens。它是推理模型，回答前先进行思考，面向笔记本和 PC 本地运行多步任务。官方称其在智能体与工具使用任务上可与 Qwen3.6-35B-A3B 竞争，后者总参数和激活参数均多约 50%。基础权重和后训练权重已在 Hugging Face 开放。
LFM2.5-24B-A2B from @liquidai is now on OpenRouter. A 24B MoE with 2B active parameters that reasons before it answers, built for agents and tool use. Liquid AI @liquidai Today we release 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱, the largest 𝙻𝙵𝙼𝟸.𝟻 family member. It is an efficient local model for laptops and PCs to work through multi-step tasks inside popular agent harnesses. > 24B total parameters, 2B active per token > Context expanded to 131K tokens > A 128K-token vocabulary > 33T pretraining tokens > It is a reasoning model > Very easy to fine-tune and customize > Performs competitively with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks. > Open base and post-trained weights on @huggingface 🤗 1/ 🔗 View Quoted Tweet 💬 3 🔄 5 ❤️ 38 👀 4084 📊 6 ⚡
