Liquid AI news and milestones

LFM2.5-DSpark speculative decoding delivers up to 3.2x faster inference on GPU and 2.9x on-device. Day-one llama.cpp and SGLang support.

What became standard practice after vibe-coding a production-grade BPE tokenizer trainer without reading a single line of code.

Liquid AI released LFM2.5-VL-3B on August 12, 2026, a 3.1-billion-parameter open-weight vision-language model built to run on phones, laptops, and single GPUs rather than in a da…

LFM2.5-VL-3B is Liquid AI's most capable vision-language model, with grounding, screen understanding, and function calling, fast enough to run on-device.

One test found it performed 3.7 times faster than DeepSeek-V4-Flash, the model has skyrocketed to the top of OpenRouter since its release last week.

Setapp operator MacPaw partners with Liquid AI to build an on-device AI stack for its Eney assistant for fast, private work.

Liquid Foundation Models and MacPaw's own AI stack will create the first Mac assistant with on-device intelligence, persistent memory and native task execution.

2.6B dense model trained for agentic workloads, with a 128K context window and native tool calling for on-device agents

We're on a journey to advance and democratize artificial intelligence through open source and open science.

LFM2.5-2.6B is an on-device agentic model that plans, calls tools, and runs multi-step tasks at 220 tok/s in under 2.5 GB. Open weights on Hugging Face.

A Blog post by Liquid AI on Hugging Face

How we doubled LFM2.5-8B-A1B's tokenizer from 65K to 128K in place with no retraining from scratch, making Hindi, Vietnamese, and Thai up to 3.7 faster on-device.

Antidoom uses Final Token Preference Optimization to cut doom-loop rates in reasoning models by up to 90%, with minimal disruption to model behavior.

This web app runs a GPU-accelerated AI model right in your browser, letting you type a text prompt (and optionally upload an image) to create a new picture instantly. You only ne…

LFM2.5-230M proves that while 3-billion-parameter models like VibeThinker are solving advanced calculus, a 230-million-parameter model is the superior, highly optimized choice fo…

Meet LFM2.5-230M, Liquid AI's smallest model yet: a fast, open-weight foundation model for fine-tuning, edge deployment, tool use, and data extraction.

Explore Liquid AI's LFM2.5-8B-A1B, an on-device Mixture of Experts model with 128K context, fast tool calling, and strong AI benchmarks.

Today, we release LFM2.5-VL-450M, an improved version of LFM2-VL-450M with grounding capabilities, better instruction following, and function calling support. The result is a com…

Today, we release LFM2.5-350M. Agentic loops at 350M parameters.A 350M model trained for reliable data extraction and tool use, where models at this scale typically struggle.500M…

Today, we're releasing LFM2.5-350M, an improved version of our 350M model with additional pre-training (from 10T to 28T tokens) and large-scale reinforcement learning. Built on t…

Building a local AI agent sounds great until you try to use one all day. The hard part isn't getting a model to understand you, it's getting it to choose the right tool and do it…

Today, we release an early checkpoint of LFM2-24B-A2B, our largest LFM2 model. This sparse Mixture of Experts (MoE) model has 24 billion total parameters with 2 billion active pe…

Liquid today announced a breakthrough in AI model training and customization that enables 350M-2.6B parameter foundation models (

We're launching Liquid Nanos - a family of 350M-2.6B parameter foundation models that deliver frontier-model quality on specialized, agentic tasks while running directly on phone…

We're excited to announce LFM2-2.6B, the newest and currently largest model in our Liquid Foundation Model 2 series. Building on our 350M, 700M, and 1.2B models, LFM2-2.6B proves…

Type a natural-language request and the app will automatically call the appropriate online tools-like calculators, search or data fetchers-to handle it. The results are gathered…

Today, we release LFM2-VL, our first series of vision-language foundation models. These multimodal models are designed for low-latency and device-aware deployment. LFM2-VL extend…

This web app lets you type a question or request and the AI model answers it, automatically calling helpful tools like calculators or web searches when needed. Just enter plain t…

Models with equal cache sizes use memory differently. Effective State-Size (ESS) measures memory utilization.

The world's best-in-class English, Arabic, and Japanese model, native in French, German, and Spanish, optimized to be the substrate for private enterprise chat, code, fast instru…

At Yahoo Finance, you get free stock quotes, up-to-date news, portfolio management resources, international market data, social interaction and mortgage rates that help you manag…

We are pleased to announce our Series A round of financing with AMD Ventures as strategic lead.

Boston's hottest artificial intelligence startup, Liquid AI, introduced a global audience to its unique and potentially revolutionary approach to AI.

Today, we report advances in automated neural network architecture design and customization. We developed algorithms for the synthesis of tailored architectures (STAR), based on…

The STAR framework from Liquid AI uses evolutionary algorithms and a numerical encoding system to balance quality and efficiency in AI models.

Artificial intelligence startup and MIT spinoff Liquid AI Inc. today launched its first set of generative AI models, and they're notably different from competing models because t…

The MIT spinoff is making its software accessible to anyone over the internet for further testing.

Announcing the first series of Liquid Foundation Models (LFMs) - a new generation of generative AI models that achieve state-of-the-art performance at every scale, while maintain…

Liquid AI, a new MIT spinoff, has raised nearly $40 million in a seed round to build an entirely new type of AI called a liquid neural network.

We are thrilled to unveil Liquid AI - an MIT spinoff - to forge the future of AI with enterprise-grade foundation models that are capable, private, and reliable.