# agentsite-static-version=f24ffa44-65a0-4796-9d93-d60b6e8da4cd
# www.liquid.ai

> AI-optimized mirror of www.liquid.ai containing 141 pages totalling 134,267 words of clean markdown content, structured data, and semantic HTML. Original source: https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models. Last updated: 2026-09-25T16:07:37.116Z.

Each page is available as HTML (with JSON-LD structured data) and Markdown (text-only, ideal for LLMs and RAG).

## Homepage

- [Liquid AI | Device-native foundation models.](/content/site-root.html): Liquid AI builds efficient Liquid Foundation Models (LFMs) for on-device, edge, and cloud AI with low latency, privacy, and hardware-aware deployment. (299 words)

## Articles & Blog Posts

- [blog/lfm2-5-8b-a1b/index.html](/content/blog/lfm2-5-8b-a1b/index.html) (1,417 words)
- [research/lfm-1b-math-can-small-models-be-concise-reasoners/index.html](/content/research/lfm-1b-math-can-small-models-be-concise-reasoners/index.html) (2,365 words)
- [liquid-foundation-models/index.html](/content/liquid-foundation-models/index.html) (2,035 words)
- [Research | Liquid AI](/content/research/index.html): Liquid Labs: papers, benchmarks, and technical reports driving the science behind Liquid Foundation Models. (1,411 words)
- [research/automated-architecture-synthesis-via-targeted-evolution.html](/content/research/automated-architecture-synthesis-via-targeted-evolution.html) (1,282 words)
- [enterprise-solutions/index.html](/content/enterprise-solutions/index.html) (357 words)
- [research/convolutional-multi-hybrids-for-edge-devices/index.html](/content/research/convolutional-multi-hybrids-for-edge-devices/index.html) (1,166 words)
- [LFM License | Liquid AI | Liquid AI](/content/lfm-license/index.html): The LFM Open License v1.0 sets out the terms under which you can use our models. This page includes a summary of this license, a FAQ, and the full license text, along with a comparison to Apache 2.0. (2,402 words)
- [press/liquids-edge-ai-platform-leap-expands-support-to-laptops-with-best-in-class-performance-on-amd-ryzen-tm-and-ryzen-ai-tm-processors.html](/content/press/liquids-edge-ai-platform-leap-expands-support-to-laptops-with-best-in-class-performance-on-amd-ryzen-tm-and-ryzen-ai-tm-processors.html) (871 words)
- [research/liquid-neural-networks-research/index.html](/content/research/liquid-neural-networks-research/index.html) (508 words)
- [company/news/index.html](/content/company/news/index.html) (290 words)
- [Research | Liquid AI](/content/news/research/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (1,833 words)
- [solutions/index.html](/content/solutions/index.html) (357 words)
- [ShieldFlow Terms of Service | Liquid AI | Liquid AI](/content/shieldflow-terms/index.html): The Terms of Service governing your access to and use of the Liquid AI ShieldFlow application. (1,960 words)
- [news/engineering/index.html](/content/news/engineering/index.html) (88 words)
- [press/liquid-ai-announces-multi-year-partnership-with-shopify-to-bring-sub-20ms-foundation-models-to-core-commerce-experiences.html](/content/press/liquid-ai-announces-multi-year-partnership-with-shopify-to-bring-sub-20ms-foundation-models-to-core-commerce-experiences.html) (543 words)
- [Company news (Page 2) | Liquid AI](/content/news/company-news/page/2/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (391 words)
- [Alexander Amini | Team | Liquid AI](/content/team/alexander-amini/index.html): Dr. Alexander Amini is the Co-founder and the CSO of Liquid AI as well as an AI Scientist at the Massachusetts Institute of Technology (MIT). (100 words)
- [Jeffrey Li | Team | Liquid AI](/content/team/jeffrey-li/index.html): Jeffrey Li is the Chief Operating Officer of Liquid AI, where he leads cross-functional execution across Product, Engineering, and Research, translating company priorities into delivery roadmaps, allocating resources, and continuously measuring and improving execution quality.  (108 words)
- [Mathias Lechner | Team | Liquid AI](/content/team/mathias-lechner/index.html): Mathias Lechner is the Co-founder and CTO of Liquid AI, as well as a Research Affiliate at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT, where he collaborates with Prof. Daniela Rus.  (217 words)
- [Ramin Hasani | Team | Liquid AI](/content/team/ramin-hasani/index.html): 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).  (152 words)
- [request-access/index.html](/content/request-access/index.html) (17 words)
- [Models | Liquid AI](/content/news/models/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (316 words)
- [Daniela L. Rus | Team | Liquid AI](/content/team/daniela-l-rus/index.html): Daniela L. Rus is the Co-founder of Liquid AI. She serves as Director of MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Andrew and Erna Viterbi Professor in MIT's Department of Electrical Engineering and Computer Science. (92 words)
- [Agentic AI | Solutions | Liquid AI](/content/agentic-ai/index.html): Proactive edge AI agents optimized for any device: capable, trustworthy agents that act locally without runaway cloud costs. (382 words)
- [News (Page 4) | Liquid AI](/content/news/page/4/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (321 words)
- [Careers | Liquid AI](/content/careers/index.html): Explore open roles at Liquid AI and help build a category-defining technology: efficient, general-purpose AI at every scale. (723 words)
- [Company | Liquid AI](/content/company/index.html): Liquid AI is building foundation models that respect privacy, performance, and domain, from architecture to inference. (550 words)
- [News (Page 3) | Liquid AI](/content/news/page/3/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (400 words)
- [ShieldFlow: On-Device AI Privacy | Liquid AI](/content/shieldflow/index.html): ShieldFlow redacts PII, API keys, and sensitive data before prompts reach AI services. It runs locally so your data stays on device. (1,001 words)
- [News (Page 2) | Liquid AI](/content/news/page/2/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (281 words)
- [oct-23rd-2024-live-stream/index.html](/content/oct-23rd-2024-live-stream/index.html) (23,440 words)
- [News (Page 5) | Liquid AI](/content/news/page/5/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (366 words)
- [Events | Liquid AI](/content/events/index.html): Where to find Liquid AI in person: the conferences, summits, and forums we’re attending in 2026. (243 words)
- [Hackathon Rules & Terms | Liquid AI | Liquid AI](/content/hackathons/terms/index.html): Official rules and terms for Liquid AI hackathons — eligibility, entry, submissions, judging, prizes, intellectual property, and liability. (936 words)
- [Terms & Conditions | Liquid AI | Liquid AI](/content/terms-conditions/index.html): Explore Liquid AI's Terms & Conditions for guidelines on website usage, content creation, account management, and legal responsibilities. (946 words)
- [Automotive | Solutions | Liquid AI](/content/automotive/index.html): On-device AI for automakers: real-time, personalized in-car assistants that run on the vehicle's existing CPUs and NPUs. (357 words)
- [Research (Page 2) | Liquid AI](/content/news/research/page/2/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (445 words)
- [News (Page 6) | Liquid AI](/content/news/page/6/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (202 words)
- [Financial services | Solutions | Liquid AI](/content/financial-services/index.html): Real-time financial AI for fraud, service, and risk workflows: low latency, private, and efficient on your infrastructure. (410 words)
- [News | Liquid AI](/content/news/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (296 words)
- [Privacy Policy | Liquid AI | Liquid AI](/content/privacy-policy/index.html): Liquid AI's Privacy Policy: Learn about data security, your rights, and account management. (628 words)
- [Industrial & Robotics | Solutions | Liquid AI](/content/industrial-robotics/index.html): Powerful AI for autonomy and automation, purpose-built to run anywhere. Bridging the gap between sophisticated AI and industrial reality. (376 words)
- [Hackathons | Liquid AI](/content/hackathons/index.html): Build on-device AI with Liquid Foundation Models. Upcoming Liquid AI hackathons and recaps from past builds. (478 words)
- [E-commerce | Solutions | Liquid AI](/content/e-commerce/index.html): On-device AI for e-commerce: private, personalized shopping experiences that run locally on every customer's device. (363 words)
- [Healthcare & Life sciences | Solutions | Liquid AI](/content/healthcare/index.html): Private healthcare AI for clinical and pharma workflows: low-latency, reliable, and deployable on private infrastructure. (361 words)
- [FAQ | Liquid AI](/content/faq/index.html): Frequently asked questions about Liquid AI's models and products. (570 words)
- [Request access | Liquid AI](/content/request-access-mmai/index.html): Liquid AI is an efficiency-first foundation model company. We build highly capable, compute-optimized models that bring intelligence to any device and medium of choice. (45 words)
- [Defense | Solutions | Liquid AI](/content/defense/index.html): On-device AI for the tactical edge: threat detection, autonomous navigation, and real-time ISR inside existing SWaP budgets. (355 words)
- [Consumer electronics | Solutions | Liquid AI](/content/consumer-electronics/index.html): On-device AI for consumer electronics: efficient multimodal SLMs for distinctive features across your device portfolio. (337 words)
- [Pricing | Liquid AI](/content/pricing/index.html): Use Liquid Foundation Models commercially for free until $10M in annual revenue. See licensing terms and enterprise options. (506 words)
- [Liquid Foundation Models | Liquid AI](/content/models/index.html): Explore Liquid Foundation Models (LFMs): efficient text, vision, audio, and nano models built for on-device, edge, and cloud AI deployment. (271 words)
- [Company news | Liquid AI](/content/news/company-news/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (149 words)
- [Engineering Blog | Liquid AI](/content/engineering/index.html): Engineering deep-dives from Liquid AI: how we build, train, and ship efficient foundation models. (93 words)
- [Connect | Liquid AI](/content/connect/index.html): Get in touch with the Liquid AI team. Tell us about your use case and we'll be in touch. (71 words)
- [Case studies | Liquid AI](/content/news/case-studies/index.html): Latest from Liquid AI: model launches, customer stories, company news, and research. (51 words)
- [Liquid Apollo](/content/apollo/index.html): Interact with AI instantly, securely, and locally. Liquid Apollo gives you a low-latency, cloud-free playground in your pocket with no barrier to entry. (23 words)
- [Request access to ShieldFlow | Liquid AI](/content/request-access-shieldflow/index.html): Liquid AI is an efficiency-first foundation model company. We build highly capable, compute-optimized models that bring intelligence to any device and medium of choice. (40 words)
- [Shoutouts | Liquid AI](/content/shoutouts/index.html): What people are saying about Liquid AI and its models. (6,400 words)
- [LFM2.5-VL-DSpark: Accelerating vision-language models on edge and beyond | Blog | Liquid AI](/content/blog/lfm2-5-vl-dspark/index.html): LFM DSpark accelerates local ai vision-language inference up to 3.13× on edge devices and 2.66× on GPUs with no quality loss. (2,522 words, 2026-09-24)
- [Liquid AI, in Collaboration with Qualcomm Technologies, Brings Personal AI Context to Devices Powered by Snapdragon | Blog | Liquid AI](/content/blog/liquid-context-qualcomm/index.html): Liquid AI, in Collaboration with Qualcomm Technologies, Brings Personal AI Context to Devices Powered by Snapdragon (914 words, 2026-09-23)
- [Setting the Frontier of Aging Biology with Liquid Foundation Models | Blog | Liquid AI](/content/blog/longevitybench/index.html): LFM2-1.2B-Longevity and LFM2-2.6B-Longevity interpret clinical, methylation, transcriptomic, proteomic, and genetic data, matching far larger models. (1,361 words, 2026-09-17)
- [Introducing Pipette: A benchmarking suite for on-device intelligence | Blog | Liquid AI](/content/blog/pipette-on-device-ai-benchmarking-by-liquid-ai/index.html): Meet Pipette, an open-source platform for reproducible on-device AI benchmarks across models, quantization, runtimes and hardware. (1,979 words, 2026-08-24)
- [LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBook | Blog | Liquid AI](/content/blog/lfm2-5-dspark.html): 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. (1,942 words, 2026-08-20)
- [LFM2.5 Q4_0: Quantization-Aware Distillation for Edge Deployment | Blog | Liquid AI](/content/blog/qad/index.html): LFM2.5 Q4_0 GGUF checkpoints trained with Quantization Aware Distillation (QAD), recovering up to 97% of BF16 accuracy for faster edge deployment. (522 words, 2026-08-19)
- [Designing Loops for Production-Grade Work | Blog | Liquid AI](/content/blog/agent-loops/index.html): What became standard practice after vibe-coding a production-grade BPE tokenizer trainer without reading a single line of code. (2,366 words, 2026-08-18)
- [LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge | Blog | Liquid AI](/content/blog/lfm2-5-vl-3b/index.html): 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. (1,582 words, 2026-08-12)
- [MacPaw Partners with Liquid AI to Bring On-Device AI to Millions of Mac Users | Blog | Liquid AI](/content/blog/macpaw-partners-liquid-ai-on-device-ai-mac-users/index.html): MacPaw partners with Liquid AI to bring private, fast, on-device AI to millions of Mac users, starting with the Eney assistant for macOS. (625 words, 2026-08-05)
- [LFM2.5-2.6B: Deploy Agents Everywhere | Blog | Liquid AI](/content/blog/lfm2-5-2-6b/index.html): 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. (1,507 words, 2026-08-04)
- [LFM2.5-Encoders: Fast at Long Context, Even on CPU | Blog | Liquid AI](/content/blog/lfm2-5-encoders/index.html): Open-weight bidirectional encoders for classification, routing & NLU. LFM2.5-Encoder-230M & 350M deliver 8K context and fast CPU inference for edge and on-prem. (1,561 words, 2026-07-28)
- [Tokenizer Expansion: Upgrading a Model's Tokenizer in Place | Blog | Liquid AI](/content/blog/tokenizer-expansion/index.html): How we doubled LFM2.5-8B-A1B's tokenizer from 65K to 128K in place, making Hindi, Vietnamese, and Thai up to 3.7× faster on-device. (1,645 words, 2026-07-21)
- [Reducing Doom Loops with Final Token Preference Optimization | Blog | Liquid AI](/content/blog/antidoom/index.html): Antidoom uses Final Token Preference Optimization to cut doom-loop rates in reasoning models by up to 90%, with minimal disruption to model behavior. (1,771 words, 2026-07-07)
- [IFStruct: Measuring structured-output compliance | Blog | Liquid AI](/content/blog/ifstruct-v1-0.html): Meet IFStruct, an open-source benchmark for testing LLM structured output, schema following, and JSON/YAML validity under realistic user prompts. (1,653 words, 2026-06-30)
- [LFM2.5-230M: Built to Run Anywhere | Blog | Liquid AI](/content/blog/lfm2-5-230m/index.html): 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. (838 words, 2026-06-25)
- [LFM2.5 Retrievers: Bi-directional LFMs for Fast Multilingual Search | Blog | Liquid AI](/content/blog/lfm2-5-retrievers/index.html): Explore Liquid AI’s LFM2.5 Retrievers: 350M multilingual embedding and ColBERT models for fast, accurate cross-lingual search across 11 languages. (1,137 words, 2026-06-18)
- [LFM2.5-8B-A1B: An Even Better On-Device Mixture of Experts | Blog | Liquid AI](/content/blog/lfm2-5-8b-a1b/index-2.html): 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., Use Loom to record quick videos of your screen and cam. Explain anything clearly and easily – and skip the meeting. An essential tool for hybrid workplaces. (1,412 words, 2026-05-28)
- [Accelerating Vision-Language Model Deployment for Automotive AI | Case study | Liquid AI](/content/use-cases/accelerating-vision-language-model-deployment-for-automotive-ai.html): Deploy 10x faster in-car AI with Liquid’s optimized vision-language models—no hardware upgrades needed. (143 words, 2026-05-14)
- [Optimizing Vision-Language Models for Product Cataloging | Case study | Liquid AI](/content/use-cases/optimizing-vision-language-models-for-product-cataloging.html): Liquid’s vision-language models cut cataloging time by 65% while delivering higher accuracy and lower costs. (97 words, 2026-05-14)
- [Powering Custom Translation Models for Smartphones | Case study | Liquid AI](/content/use-cases/powering-custom-translation-models-for-smartphones/index.html): Liquid’s smartphone translation models deliver real-time performance with 50% less memory—no cloud needed. (116 words, 2026-05-14)
- [Real-Time Speed, Scale, and Precision in Fraud Prevention: +~$230M Fraud Detected Annually | Case study | Liquid AI](/content/use-cases/real-time-speed-scale-and-precision-in-fraud-prevention-230m-fraud-detected-annually.html): We partnered closely with the client to develop a custom model powered by the Liquid Foundation Model (LFM) backbone, designed to deliver speed, scalability, and stronger detection performance. (429 words, 2026-05-14)
- [Scaling Synthetic Video Generation Without Cloud Bottlenecks | Case study | Liquid AI](/content/use-cases/scaling-synthetic-video-generation-without-cloud-bottlenecks.html): Cut cloud costs by 70% with Liquid’s local synthetic video models—built for speed and scale. (94 words, 2026-05-14)
- [Liquid AI and Mercedes-Benz partner to scale embedded in-car intelligence | Press | Liquid AI](/content/press/liquid-ai-and-mercedes-benz-partner-to-scale-embedded-in-car-intelligence.html): Liquid AI announces multi-year partnership with Mercedes-Benz (509 words, 2026-04-23)
- [LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud | Blog | Liquid AI](/content/blog/lfm2-5-vl-450m/index.html): 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 compact model that turns image streams into structured, actionable outputs in real time, even on edge hardware. (828 words, 2026-04-08)
- [LFM2.5-350M: No Size Left Behind | Blog | Liquid AI](/content/blog/lfm2-5-350m-no-size-left-behind/index.html): 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 the LFM2 architecture, it delivers exceptionally fast inference and runs everywhere, from cloud GPUs to cheap CPUs.  (1,557 words, 2026-03-31)
- [No Cloud, No Waiting: Tool-Calling Agents on Consumer Hardware with LFM2-24B-A2B | Blog | Liquid AI](/content/blog/no-cloud-tool-calling-agents-consumer-hardware-lfm2-24b-a2b.html): 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 fast enough that the experience feels interactive. In regulated industries, the bar is even higher: the agent also cannot send your data anywhere to do it. This is where LFM2-24B-A2B shines: it’s designed for tool dispatch on local consumer hardware, where latency, memory and privacy aren’t abstract constraints; they decide whether your agent is a product or a demo. (1,064 words, 2026-03-05)
- [Liquid AI and Insilico Medicine Announce Strategic Partnership Delivering Lightweight Scientific Foundation Models for Drug Discovery | Press | Liquid AI](/content/press/liquid-ai-insilico-medicine-partnership/index.html): Liquid AI and Insilico Medicine today announced a partnership that creates lightweight scientific foundation models for pharmaceutical research. The collaboration has produced LFM2-2.6B-MMAI, available now – a single checkpoint trained to perform at state-of-the-art levels across multiple drug discovery subdomains, not a patchwork of separate point models.  (729 words, 2026-03-03)
- [LFM2-24B-A2B: Scaling Up the LFM2 Architecture | Blog | Liquid AI](/content/blog/lfm2-24b-a2b/index.html): 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 per token, showing that the LFM2 architecture scales effectively to larger sizes. (915 words, 2026-02-24)
- [From Cloud to AI PC: Launching Liquid’s Largest LFM2 Model Alongside Our Growing Partner Ecosystem | Blog | Liquid AI](/content/blog/lfm2-24b-a2b-from-cloud-to-ai-pc/index.html): Today, we are completing the LFM2 family with the launch of our most capable model yet: LFM2-24B-A2B. While our Technical Blog dives into the architectural specs, the real story is its versatility. This model is designed to run everywhere, from massive cloud clusters to the next generation of AI PCs and mobile devices. (966 words, 2026-02-24)
- [LFM2.5-1.2B-Thinking: On-Device Reasoning Under 1GB | Blog | Liquid AI](/content/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb/index.html): Today, we are releasing LFM2.5-1.2B-Thinking, a reasoning model that runs entirely on-device. It fits within 900 MB of memory on a phone and delivers both the fastest inference speed and the best quality for its size. What required a data center two years ago now runs offline in your pocket. (1,557 words, 2026-01-20)
- [The Future of AI Isn't Just Bigger; It’s Specialized and Hybrid | Blog | Liquid AI](/content/blog/the-future-of-ai-isnt-just-bigger-its-specialized-and-hybrid.html): For the last few years, the AI narrative has been dominated by a 'bigger is better' philosophy. We’ve watched parameters balloon into hundreds of billions or even trillions, driven by the assumption that raw scale is the only path to intelligence. While scale delivers capability, it also creates a deployment bottleneck. A shift is now happening because massive models are too heavy for every task. The future of AI, particularly in Agentic Workflows, is not about choosing one size, it’s about a hybrid approach." (1,411 words, 2026-01-14)
- [The Future of Meeting Summarization: Local, Fast, Private, and Fully Secure | Blog | Liquid AI](/content/blog/the-future-of-meeting-summarization-local-fast-private-and-fully-secure.html): This week at CES, our Liquid AI team and partners at AMD are showcasing the future of on-device intelligence with lightning fast, reliable and entirely secure meeting summarization, run locally on AMD hardware.Meetings capture critical business intelligence: decisions, commitments, customer insights, action items, and often, confidential and sensitive information. Most AI summarization tools require sending that data to the cloud, introducing latency, unpredictable costs, and real security and compliance risks. (1,106 words, 2026-01-07)
- [Introducing LFM2.5: The Next Generation of On-Device AI | Blog | Liquid AI](/content/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai.html): Today, we're excited to announce the LFM2.5-1.2B model family, our most capable release yet for edge AI deployment. It builds on the LFM2 device-optimized architecture and represents a significant leap forward in building reliable agents on the edge. (1,398 words, 2026-01-05)
- [Introducing Liquid Labs: Defining the Frontier of Intelligence through Innovation | Blog | Liquid AI](/content/blog/introducing-liquid-labs-defining-the-frontier-of-intelligence-through-innovation.html): Research has been core to Liquid AI from the start. Today, we’re giving that work a formal name: Liquid Labs, our team driving fundamental breakthroughs in the science of building intelligent, personalized, and adaptive machines. (310 words, 2025-12-04)
- [LFM2: Advancing Open Science in AI | Blog | Liquid AI](/content/blog/lfm2-advancing-open-science-in-ai/index.html): At Liquid AI, we believe that scientific progress accelerates when knowledge is shared openly. Today, we're pleased to release the full technical report for LFM2, our second generation of Liquid Foundation Models. (153 words, 2025-12-01)
- [Hack the Edge San Francisco: Accelerating What’s Possible with Edge-Native AI | Blog | Liquid AI](/content/blog/hack-the-edge-san-francisco-accelerating-whats-possible-with-edge-native-ai.html): This weekend, more than 70 engineers, researchers, and builders converged at Liquid AI’s San Francisco office for Hack the Edge, a 48-hour sprint co-hosted with AMD. Equipped with AMD mini-PCs powered by Ryzen™ AI, Liquid Foundation Models and the ROCm™ stack, teams built, fine-tuned, and deployed edge-native applications entirely on-device. (328 words, 2025-11-18)
- [Liquid AI Announces Multi‑Year Partnership with Shopify to Bring Sub‑20ms Foundation Models to Core Commerce Experiences | Blog | Liquid AI](/content/blog/liquid-ai-announces-multi-year-partnership-with-shopify-to-bring-sub-20ms-foundation-models-to-core-commerce-experiences.html): Shopify’s multimillion-dollar deal licenses Liquid AI’s low-latency LFMs for search and a co-built generative recommender that beat prior systems in tests. (538 words, 2025-11-13)
- [LFM2-ColBERT-350M: One Model to Embed Them All | Blog | Liquid AI](/content/blog/lfm2-colbert-350m-one-model-to-embed-them-all/index.html): Today, we release LFM2-ColBERT-350M, a late interaction retriever with excellent multilingual performance. It allows you to store documents in one language (for example, a product description in English) and retrieve them in many languages (e.g., a query in Spanish, German, French, etc.) with high accuracy. Thanks to the efficient LFM2 backbone, it also benefits from extremely fast inference speed, on par with models that are 2.3 times smaller. (702 words, 2025-10-28)
- [Agentic Robotics at the Edge: Liquid AI, AMD, and Robotec.ai Demonstrate LFM2-VL-3B in Embedded Autonomy | Blog | Liquid AI](/content/blog/agentic-robotics-at-the-edge-liquid-ai-amd-and-robotec-ai-demonstrate-lfm2-vl-3b-in-embedded-autonomy.html): Together, Liquid AI, AMD, and Robotec.ai have deployed compact foundation models for autonomous agentic robotics: showcasing a specialized 3-billion parameter Liquid vision-language model (LFM2-VL-3B), running efficiently on AMD Ryzen™ AI processors, to enable real-time multimodal perception and decision-making at the edge. (747 words, 2025-10-23)
- [How Liquid AI Uses ExecuTorch to Power Efficient, Flexible On-Device Intelligence | Blog | Liquid AI](/content/blog/how-liquid-ai-uses-executorch-to-power-efficient-flexible-on-device-intelligence.html): When we needed to deploy our hybrid LFM models on-device, we faced a critical challenge: existing inference engines couldn't handle the unique combination of attention and recurrent layers our architecture required. By adopting PyTorch ExecuTorch in late 2024, we achieved seamless deployment with 2× faster inference and significantly lower latency, proving that the right infrastructure choice can unlock entirely new possibilities for edge AI. (495 words, 2025-10-22)
- [LFM2-VL-3B: A New Efficient Vision-Language for the Edge | Blog | Liquid AI](/content/blog/lfm2-vl-3b-a-new-efficient-vision-language-for-the-edge.html): We’re excited to release LFM2-VL-3B, the newest and most capable addition to our family of vision LFMs (450M and 1.6B). Built on the LFM2-2.6B backbone, this 3B parameter model targets applications that require more accuracy while maintaining the speed advantage of the LFM2 architecture. It is available today on LEAP and Hugging Face. (412 words, 2025-10-22)
- [LFM2-8B-A1B: An Efficient On-device Mixture-of-Experts | Blog | Liquid AI](/content/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts.html): We are releasing LFM2-8B-A1B, our first on-device Mixture-of-Experts (MoE) with 8.3B total parameters and 1.5B active parameters per token. By activating only a sparse subset of parameters during inference, LFM2-8B-A1B delivers larger model quality with the compute of a 1.5B-class model. It trades a modest increase in memory footprint for higher quality and speed compared to dense models. This enables fast, private, low-latency applications on modern phones, tablets, and laptops. (1,508 words, 2025-10-07)
- [LFM2-Audio: An End-to-End Audio Foundation Model | Blog | Liquid AI](/content/blog/lfm2-audio-an-end-to-end-audio-foundation-model/index.html): LFM2-Audio defines a new class of audio foundation models: lightweight, multimodal, and real-time. By unifying audio understanding and generation in one compact system, it enables conversational AI on devices where speed, privacy, and efficiency matter most. With LFM2-Audio, we’re taking another step toward multimodal agents that run privately and in real time on your own devices.  (780 words, 2025-10-01)
- [Introducing Liquid Nanos — frontier‑grade performance on everyday devices | Blog | Liquid AI](/content/blog/introducing-liquid-nanos-frontier-grade-performance-on-everyday-devices.html): 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 phones, laptops, and embedded devices. In our evaluations with partners, Nanos perform competitively with models hundreds of times larger. The result: planet‑scale AI agents with cloud‑free economics.
 (1,787 words, 2025-09-25)
- [Liquid unveils “Nanos”: Extremely small foundation models that match frontier-model quality—running directly on everyday devices | Press | Liquid AI](/content/press/liquid-unveils-nanos-extremely-small-foundation-models-that-match-frontier-model-quality-running-directly-on-everyday-devices.html): Liquid today announced a breakthrough in AI model training and customization that enables 350M–2.6B parameter foundation models (“Nanos”) to deliver GPT-4o-class performance on specialized agentic tasks—while running on phones, laptops, and embedded devices. In internal and partner evaluations, Liquid Nanos perform competitively with models up to hundreds of times larger. The result: planet-scale AI agents with cloud-free economics. (713 words, 2025-09-25)
- [Introducing LFM2-2.6B: Redefining Efficiency in Language Models | Blog | Liquid AI](/content/blog/introducing-lfm2-2-6b-redefining-efficiency-in-language-models.html): Liquid AI LFM2-2.6B: a compact generative AI foundation model delivering high performance, efficient inference, and smaller-scale deployment on device. (236 words, 2025-09-23)
- [Hack02: LFMs with Eyes — What Happened, Who Won, What’s Next | Blog | Liquid AI](/content/blog/hack02-lfms-with-eyes-what-happened-who-won-whats-next.html): Liquid AI just closed Hack02, themed LFMs with Eyes. This round focused on vision-language capabilities with our new LFM2-VL model — and the results surpassed expectations. Participation doubled. Our Discord community passed 1,000 members and is now a Discoverable Community, opening the door to even more builders. (396 words, 2025-09-19)
- [Brilliant Labs Partners With Liquid AI to Bring Vision-Language Tech to Your Glasses | Press | Liquid AI](/content/press/brilliant-labs-partners-with-liquid-ai-to-bring-vision-language-tech-to-your-glasses.html): The next generation of Brilliant Labs AI glasses, Halo (available today for pre-order), will be powered by Liquid AI’s lightweight vision-language Liquid foundation models to bring fast, reliable, and fully private intelligence to wearers’ eyes. (611 words, 2025-09-15)
- [Hack01: Hack the Edge. Take the Dare. | Blog | Liquid AI](/content/blog/hack01-hack-the-edge-take-the-dare/index.html): Liquid AI’s first hackathon, Hack01, was built around a challenge: Hack the Edge. Take the Dare. We invited developers to go beyond cloud APIs and prove what’s possible when AI runs fully on-device. The results weren’t just clever demos. They were a glimpse into a new category of applications that only work when intelligence is private, fast, and resilient. (360 words, 2025-08-22)
- [Alef Education Collaborates with Liquid AI to Advance AI in Education Globally | Press | Liquid AI](/content/press/alef-education-collaborates-with-liquid-ai-to-advance-ai-in-education-globally.html): Alef Education, a global leader in education technology based in the United Arab Emirates, today announced a collaboration with Liquid AI, a leading efficient foundation models company headquartered in Cambridge, Massachusetts. This collaboration will deploy Liquid AI’s private, on-device generative models across Alef Education’s AI driven solutions, which serves more than 1.5 million students and 14,000 K–12 schools globally. (554 words, 2025-08-19)
- [From Cloud to Keyboard: LEAP Expands to Laptops with Cutting-Edge AMD Ryzen™ Power | Blog | Liquid AI](/content/blog/liquids-edge-ai-platform-leap-expands-support-to-laptops-with-best-in-class-performance-on-amd-ryzen-tm-and-ryzen-ai-tm-processors.html): Liquid AI today announces native support for AMD’s latest Ryzen™ and Ryzen AI™ processors, unlocking new performance gains on-device AI solutions.  (871 words, 2025-08-18)
- [LFM2-VL: Efficient Vision-Language Models | Blog | Liquid AI](/content/blog/lfm2-vl-efficient-vision-language-models/index.html): Today we release LFM2-VL, our first vision-language models, built for low-latency, device-aware text and image use.
 (920 words, 2025-08-12)
- [Liquid AI Hackathon Series is Live | Blog | Liquid AI](/content/blog/liquid-ai-hackathon-series-is-live/index.html): This is your shot to build with LFM2, Liquid’s next-gen small model, and LEAP, our edge-native deployment engine.
Together, they let you ship AI that’s private, real-time, and local—running where it matters most: on your device.
SLMs (Small Language Models) are perfect for fast, secure tasks that don’t belong in the cloud. LFM2 is the best-performing mobile edge model on the market, and LEAP makes it easy to drop into iOS and Android apps.
We’ve built the tech. Now we want to see what you take back. (299 words, 2025-08-02)
- [Liquid <> .txt Collaboration | Blog | Liquid AI](/content/blog/liquid-txt-collaboration/index.html): Faster and more accurate function calling on the edge with .txt’s structured outputs and Liquid Foundation Models (1,416 words, 2025-07-24)
- [Liquid AI Launches LEAP and Liquid Apollo: The Easiest Way to Build with On-Device AI | Blog | Liquid AI](/content/blog/liquid-ai-launches-leap-and-apollo-bringing-edge-ai-to-every-developer.html): Today marks a pivotal milestone in the evolution of edge AI. Liquid AI is thrilled to announce LEAP v0, our first developer-ready platform for on-device AI deployment—and Liquid Apollo, a lightweight iOS-native application built to showcase and stress-test small foundation models directly on your phone. (659 words, 2025-07-15)
- [Liquid AI Launches LEAP and Liquid Apollo: A New Era for Edge AI Deployment Begins | Press | Liquid AI](/content/press/liquid-ai-launches-leap-and-apollo-a-new-era-for-edge-ai-deployment-begins.html): New developer platform and consumer app showcase the power of reliable and efficient generative AI models deployed on local edge devices, enabling fast, private, and always-on AI experiences (487 words, 2025-07-15)
- [Introducing LFM2: The Fastest On-Device Foundation Models on the Market | Blog | Liquid AI](/content/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models.html): Today, we release LFM2, a new class of Liquid Foundation Models (LFMs) that sets a new standard in quality, speed, and memory efficiency for on-device deployment. Built on a hybrid architecture, LFM2 delivers 200% faster decode and prefill performance than Qwen3 and Gemma 3 on CPU. It also significantly outperforms models in each size class on instruction-following and function calling—the core capabilities that make LLMs reliable for building AI agents. (2,005 words, 2025-07-10)
- [Liquid AI Releases World’s Fastest and Best-Performing Open-Source Small Foundation Models | Press | Liquid AI](/content/press/liquid-ai-releases-worlds-fastest-and-best-performing-open-source-small-foundation-models.html): Next-generation edge models outperform top global competitors; now available open source on Hugging Face (595 words, 2025-07-10)
- [LFM-1B-Math: Can Small Models Be Concise Reasoners? | Blog | Liquid AI](/content/blog/lfm-1b-math-can-small-models-be-concise-reasoners/index.html): How LFM-1.3B learns concise reasoning through fine-tuning and reinforcement learning for efficient edge deployment. (2,537 words, 2025-06-24)
- [G42 and Liquid AI Partner to Deliver Private, Local and Efficient AI Solutions to Enterprises at Scale | Press | Liquid AI](/content/press/g42-and-liquid-ai-partner-to-deliver-private-local-and-efficient-ai-solutions-to-enterprises-at-scale.html): G42, the Abu Dhabi-based global technology group, and Liquid AI, a leading efficient foundation models company headquartered in Cambridge, Massachusetts, today announced that they have entered into a multifaceted commercial partnership to facilitate the creation, training and commercialization of generative AI solutions powered by Liquid Foundation Models. (368 words, 2025-06-17)
- [Liquid AI to Power On-Device Intelligence in Sharp’s Bestselling Mobile Phone | Press | Liquid AI](/content/press/liquid-ai-to-power-on-device-intelligence-in-sharps-bestselling-mobile-phone.html): Liquid AI, a foundation model company setting new standards for performance and efficiency, today announced that Sharp Corporation, a leading global technology innovator, has selected Liquid AI’s vision-language model (VLM) for integration into one of its bestselling smartphones, the SHARP AQUOS wish4. This milestone reflects a deepening technical collaboration between the two companies as they explore edge AI use cases together. (336 words, 2025-06-15)
- [Exploring Diffusion Transformer Designs via Grafting | Blog | Liquid AI](/content/blog/exploring-diffusion-transformer-designs-via-grafting.html): Grafting lets researchers edit pretrained AI architectures, cutting exploration costs while preserving model quality. (934 words, 2025-06-12)
- [How Effectively Does a Model Use Its Memory | Blog | Liquid AI](/content/blog/how-effectively-does-a-model-use-its-memory/index.html): Models with equal cache sizes use memory differently. Effective State-Size (ESS) measures memory utilization. (312 words, 2025-04-27)
- [Convolutional Multi-Hybrids for Edge Devices | Blog | Liquid AI](/content/blog/convolutional-multi-hybrids-for-edge-devices/index.html): Hyena Edge is a convolutional AI model that outperforms Transformers in quality and efficiency on edge devices. (1,127 words, 2025-04-24)
- [Liquid at ICLR 2025 | Blog | Liquid AI](/content/blog/liquid-at-iclr-2025/index.html): Liquid at ICLR 2025 (39 words, 2025-04-21)
- [Introducing LFM-7B: Setting New Standards for Efficient Language Models | Blog | Liquid AI](/content/blog/introducing-lfm-7b-setting-new-standards-for-efficient-language-models.html): Best-in-class multilingual model for private enterprise chat, code, instruction following, and agentic workflows. (1,228 words, 2025-01-20)
- [We raised $250M to scale capable and efficient general-purpose AI | Blog | Liquid AI](/content/blog/we-raised-250m-to-scale-capable-and-efficient-general-purpose-ai.html): We are pleased to announce our Series A round of financing with AMD Ventures as strategic lead. (324 words, 2024-12-13)
- [Automated Architecture Synthesis via Targeted Evolution | Blog | Liquid AI](/content/blog/automated-architecture-synthesis-via-targeted-evolution.html): STAR automates neural architecture design, creating custom AI models that outperform Transformers with fewer parameters. (1,250 words, 2024-12-01)
- [Performance at every scale and in every domain: how and where Liquid AI delivers ROI | Blog | Liquid AI](/content/blog/performance-at-every-scale-and-in-every-domain-how-and-where-liquid-ai-delivers-roi.html): Liquid AI is one of the first AI companies providing organizations with the new essentials for AI performance at scale. Capable of operating on legacy technology, on-device – even without an internet connection – all the way to the enterprise level, our suite of novel LFMs deliver scale across modalities, industries, and a variety of sequence data to drive your ambitions – and AI – forward.  (730 words, 2024-11-13)
- [Liquid AI to Unveil First Products Built on Liquid Foundation Models (LFMs) at Exclusive MIT Event | Press | Liquid AI](/content/press/liquid-ai-to-unveil-first-products-built-on-liquid-foundation-models-lfms-at-exclusive-mit-event.html): Liquid AI, an MIT spin-off and foundation model company, will unveil its first products at an exclusive launch event on October 23, 2024. These innovative products leverage Liquid Foundation Models (LFMs), offering state-of-the-art performance and more efficient, scalable solutions for on-device computing and private, on-premise enterprise use cases. (454 words, 2024-10-15)
- [Liquid Foundation Models: Our First Series of Generative AI Models | Blog | Liquid AI](/content/blog/liquid-foundation-models-our-first-series-of-generative-ai-models.html): Announcing Liquid Foundation Models: state-of-the-art local AI with a smaller memory footprint and faster inference. (1,978 words, 2024-09-30)
- [Liquid AI Introduces New Class of Foundational Models, Advancing AI Performance and Efficiency | Press | Liquid AI](/content/press/liquid-ai-introduces-new-class-of-foundational-models-advancing-ai-performance-and-efficiency.html): Liquid AI, a pioneer in artificial intelligence innovation, is set to redefine the AI landscape with the launch of its Liquid Foundational Models (LFMs). These models bring unprecedented levels of performance and efficiency to generative AI, offering powerful solutions for industries looking to harness artificial intelligence more effectively on device or on server. A technical whitepaper about model performance will be available on September 30, 2024, at 11 am EST. (445 words, 2024-09-30)
- [From Liquid Neural Networks to Liquid Foundation Models | Blog | Liquid AI](/content/blog/liquid-neural-networks-research/index.html): Explore Liquid AI research on liquid neural networks, state space models, Hyena, neural operators, and efficient AI. (451 words, 2024-09-29)
- [Liquid at ICML 2024 | Blog | Liquid AI](/content/blog/liquid-at-icml-2024/index.html): Liquid at ICML 2024 (104 words, 2024-07-09)
- [Liquid at ICLR 2024 | Blog | Liquid AI](/content/blog/liquid-at-iclr-2024/index.html): Liquid at ICLR 2024 (52 words, 2024-04-22)
- [Launch of Collaboration with Liquid AI to Develop Edge AI Solution | Blog | Liquid AI](/content/blog/launch-of-collaboration-with-liquid-ai-to-develop-edge-ai-solution.html): Aiming to enable machine learning with minimal power consumption and improve processing performance on edge devices. (440 words, 2024-02-27)
- [Liquid AI announces collaboration with Capgemini to build next-generation AI solutions for enterprises | Blog | Liquid AI](/content/blog/liquid-ai-announces-collaboration-with-capgemini-to-build-next-generation-ai-solutions-for-enterprises.html): Following Capgemini's participation in Liquid AI's successful $37.6m seed round, both companies will collaborate to bring cutting-edge AI systems, powered by Liquid Neural Networks, to help drive enterprises’ AI transformation. (473 words, 2024-01-12)
- [Liquid at NeurIPS 2023 | Blog | Liquid AI](/content/blog/liquid-at-neurips-2023/index.html): Liquid at NeurIPS 2023 (360 words, 2023-12-09)
- [Liquid AI: A New Generation of AI Models from First Principles | Blog | Liquid AI](/content/blog/new-generation-of-ai-models-from-first-principles/index.html): 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. (412 words, 2023-12-06)

## About Pages

- [contact/index.html](/content/contact/index.html) (87 words)
- [Team | Liquid AI](/content/team/index.html): Meet the people building foundation models at Liquid AI. (695 words)

## Resources

- [Full Page Index](/index.html): Browse all cached pages with rich metadata
- [About This Cache](/about.html): Methodology, technical details, and usage guidelines
- [XML Sitemap](/sitemap.xml): Machine-readable sitemap for crawler discovery
- [Robots.txt](/robots.txt): Crawler directives
- [Agentsite Network](https://agentsite.network/network.html): Public index of Agentsites and their machine-readable resources