Liquid AI 发布 38T 端侧 MoE 模型
推荐指数 62.0 NO. 015 · 2026.05.30
发布2026/05/29Score65Comments13
为什么值得看
LFM2.5-8B-A1B 将预训练数据从 12T 扩至 38T tokens，上下文窗口扩至 128K，专注消费级硬件上的工具调用链。对需要在笔记本本地跑复杂 agent 工作流的开发者，这是目前少数能兼顾性能与便携性的选择。
媒体预览
编辑判断
端侧模型赛道正在从"能跑"转向"能干活"——工具调用链的可靠性比纯聊天能力更难做，因为涉及多步状态管理和错误恢复。Liquid AI 押注这个场景很聪明，但目前端侧工具生态碎片化严重，MCP、Function Calling 各厂标准不统一，实际落地还要看和哪些 agent 框架深度集成。
对比之下，Apple 的端侧模型走封闭生态，Llama 3.2 1B/3B 虽然开源但工具调用能力弱。LFM2.5-8B-A1B 的差异化在于用 MoE 架构在 8B 激活参数下做到 1B 总参数规模的推理成本，这对电池续航敏感的移动场景很关键。做本地 AI 助手或离线自动化工具的团队可以优先测试它的工具调用延迟和准确率。
社区反馈
意见分歧 14 条评论
核心争论：小模型真实能力是否被基准测试夸大，实际体验能否兑现端侧部署承诺
This is super interesting, I'm particularly excited for this one as it may allow teams to scale this architecture for VLAs (vision language action models), and having sparser models means more real-time actions on a locally hosted model demo link for anyone that wants to try this out https:/&#x
Why does this not have (day-one) support for Ollama? The previous model is on there? Is it related to the ongoing refactor work or are people abandoning Ollama for other LLM engines?
Ollama is just llama.cpp but with their own interface ontop. Liquid does support llama.cpp, but Ollama is slow in updating its llama.cpp dependency.