Daily AI CatchupLiquid AI's LFM3 Targets Phones and Edge Devices
Liquid AI released LFM3, a family of open-weight text, vision, audio, omni, and decision models built on a shared backbone, starting around 2B parameters. The 1.7B text model outperforms its 2.6B predecessor with 35% fewer parameters, supports a 256K context window, and fine-tunes on a single GPU. The vision variant matches Gemma 4 12B across 62 benchmarks despite being far smaller.
Why it matters
💻 Developer · Open weights, single-GPU fine-tuning, and sub-1GB footprint make this a practical option for shipping on-device agents without cloud dependency.
📦 Product · Smaller, cheaper models that still hit competitive benchmarks open the door to AI features that run offline on phones and wearables.
🎨 Design · On-device models with 256K context could enable responsive, private assistant features without the latency of round-tripping to a server.
📈 Business · Cuts inference costs and removes cloud dependency for edge AI products, relevant if you're planning mobile or embedded AI features.
🤔 Just Curious · A phone-sized AI model that's smaller than its predecessor but performs better, and can run without an internet connection.
Sources: Liquid AI Launches LFM3 Models for On Device AI, LFM3: Liquid AI's new family of small, on-device multimodal models