Liquid AI has released LFM2.5-VL-3B, a lightweight vision-language model designed to run on smartphones. This open model, with 3.1 billion parameters and under 3.3GB of memory usage, supports OCR, UI recognition, and multiple languages including Japanese. Benchmarks show LFM2.5-VL-3B outperforming larger models like Gemma-4-E4B-it and Qwen3.5-4B on various processors, including those found in mobile devices, achieving a decoding speed of 20 tokens/sec on a Galaxy S26 Ultra. AI
IMPACT Enables on-device vision-language tasks, potentially accelerating mobile AI applications and reducing reliance on cloud processing.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
