2+ hour, 34+ min ago  (280+ words) Liquid's models are optimized for everything that hospitals, pharma, and medtech need. Get reliable, real-time decisions that fit your budget and keep sensitive data out of the cloud. Researchers and caregivers need their data to stay close. They also need…...

2+ hour, 2+ min ago  (31+ words) Connect Liquid AI Fill out the form below to get started. Building the fastest, most compute-efficient, and capable foundation models so intelligence can live anywhere. " 2026 Liquid AI, Inc....

2+ hour, 50+ min ago  (24+ words) Request access Liquid AI Gain access to LFM2-2. 6 B-MMAI, a small model that achieves cloud-scale performance while operating entirely on private infrastructure....

2+ hour, 53+ min ago  (224+ words) Ramin Hasani is the Co-founder and CEO of Liquid AI and a machine learning Scientist at the Computer Science and Artificial Intelligence Lab (CSAIL), Massachusetts Institute of Technology (MIT). In his research role at MIT, he focuses on developing robust…...

2+ day, 6+ hour ago  (344+ words) The base (LFM2. 5-230 M-Base) and post-trained (LFM2. 5-230 M) models are available today on Hugging Face. Check out our docs on how to run and fine-tune them locally. The model was pre-trained for 19 T tokens, including a 32 K context extension phase. We apply a…...

4+ week, 1+ day ago  (745+ words) Today, we're releasing LFM2. 5-8 B-A1 B, an edge model built for fast, reliable tool calling on consumer hardware. It builds on our LFM2-8 B-A1 B release from October 2025, with an expanded 128 K context window, scaled-up pretraining (from 12 T to 38 T tokens), and large-scale reinforcement…...

2+ mon, 3+ day ago  (157+ words) The partnership aims to make the driver/vehicle relationship even more natural and effortless. Built on the in-house developed Mercedes-Benz Operating System (MB. OS) software architecture, the collaboration creates a path to a first production deployment for advanced speech technology,…...

2+ mon, 2+ week ago  (413+ words) Today, we release LFM2. 5-VL-450 M, an improved version of LFM2-VL-450 M with grounding capabilities, better instruction following, and function calling support. The result is a compact model that turns image streams into structured, actionable outputs in real time, even on edge hardware....

2+ mon, 3+ week ago  (447+ words) The base (LFM2. 5-350 M-Base) and post-trained (LFM2. 5-350 M) models are available today on Hugging Face, LEAP, and our Playground. Check out our docs on how to run and fine-tune them locally. We evaluated LFM2. 5-350 M across ten benchmarks spanning core capabilities and applied tasks....

4+ mon, 3+ day ago  (553+ words) Today, we release an early checkpoint of LFM2-24 B-A2 B, our largest LFM2 model. This sparse Mixture of Experts (Mo E) model has 24 billion total parameters with 2 billion active per token, showing that the LFM2 architecture scales effectively to larger sizes. With this release, the…...