Liquid AI Launches LFM2.5-VL-3B: A 2GB On-Device Vision Model with Massive Grounding Gains
Fun-Meaning-6474 · reddit · 2026-08-13
Liquid AI has released LFM2.5-VL-3B, a 3.1B parameter vision model roughly 2GB in size, optimized for on-device deployment on phones.
- Grounding Improvements: The model shows a massive leap in object grounding compared to its predecessor. Its score on the ScreenSpot-v2 desktop benchmark jumped from 6 to 78.7.
- Real-world Test: A developer tested the model locally on an iPhone 17 to recognize a Minecraft Steve toy. While inference took about 2 minutes and 31 seconds, it successfully identified the character and provided a detailed, part-by-part description leveraging its enhanced grounding skills.
Related event: LiquidAI Releases 3B Edge Vision Model LFM2.5-VL-3B(9 posts)→
More from Models
- LiquidAI Launches LFM2.5-VL-3B: Lightweight VLM Runs Tool Calling Locally — helloiamleonie · 2026-08-13
- Gemini Robotics 2 Demonstrates Unscripted 3D Whole-Body Generalization — keerthanpg · 2026-08-13
- Vision+Logic Benchmark: Gemini Leads, GPT Stagnates — Afinetheorem · 2026-08-13
- Gemini Robotics Tested: Humanoid Robot Shows Human-like Motor Generalization — keerthanpg · 2026-08-13
- Cohere Models Gain MLX Support for Fast Local Execution on Apple Devices — cohere · 2026-08-13
- LMSYS Arena Introduces AutoEval, a Reward Model for Automated Evaluation — arena · 2026-08-13