What Liquid AI does

Liquid AI is a 2023 MIT CSAIL spinoff headquartered in Cambridge, Massachusetts, founded by Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela Rus, the original researchers behind liquid neural networks. The company develops Liquid Foundation Models (LFMs), a family of generative AI foundation models built on a proprietary architecture combining Linear Input-Varying (LIV) systems with grouped query attention, representing a non-transformer alternative designed for efficient on-device deployment. Models range from 350M to 24B parameters across three generations (LFM1, LFM2, LFM2.5) with multimodal variants covering text, vision, audio, retrieval, and scientific domains. The company claims up to 1,000x smaller model footprint than frontier LLMs, with the CompreSSM compression technique achieving 74% FLOP reduction while retaining 96.5-99.2% of quality.

The company operates a multi-channel go-to-market: free open-weight distribution under the LFM Open License for organizations under $10M annual revenue and research use, with commercial licensing above that threshold; the LEAP developer platform for customization and deployment on edge hardware (AMD Ryzen, Qualcomm); OEM embedded partnerships integrating LFMs into partner products; multi-year strategic enterprise contracts with companies including Mercedes-Benz (in-car AI for MBUX), Shopify (commerce search), and Insilico Medicine (drug discovery); and a direct-to-consumer mobile app, Liquid Apollo, showcasing on-device AI. As of 2025, Liquid AI reported $13.2M in annual recurring revenue, supported by $287.6M in cumulative funding across a $37.6M seed round (December 2023) and a $250M Series A led by AMD Ventures (December 2024) at a $2.4B post-money valuation.

The company maintains active research collaborations with MIT, Max Planck Institute, and ETH Zurich, and has built a distribution ecosystem spanning cloud marketplaces (Together AI, Modal, Amazon Bedrock), edge runtimes (Ollama, LM Studio, ExecuTorch, llama.cpp, vLLM, SGLang), and silicon partners (AMD, Qualcomm). The 2,000+ developer community, presence at top ML research conferences (ICML, ICLR, NeurIPS), and sustained coverage in tier-one media (WSJ, Wired, Bloomberg, Forbes, McKinsey) support both talent acquisition and demand generation. As of May 2026, Apple is reportedly in acquisition discussions with the company.

Name

Liquid AI

Legal name

Liquid AI, Inc.

Website

https://liquid.ai

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

101–250 employees

Short description

Liquid AI, a 2023 MIT CSAIL spinoff, develops compact Liquid Foundation Models (LFMs) using a proprietary liquid neural network architecture for efficient on-device AI deployment. The company serves enterprise customers across automotive, commerce, healthcare, and consumer electronics via multi-year strategic contracts, open-weight distribution, and OEM embedded partnerships.

Ownership category

akta.pro rank

Product category

AI Foundation Models

NAICS

Software Publishers (5132)

SIC

Services-Commercial Physical & Biological Research (8731)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM), Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO), Agents & Autonomous Workflows (Tool Use, Planning, Multi-Agent) (HDAAACAF), Model Compression & Optimization (Quantization, Distillation, Pruning) (HDAAACAL), AI/ML Engineering & Model Development Services (BPAEAHAF)

Keywords

• Foundation models
• Edge AI deployment
• On-device AI
• AI developer platform
• Model licensing

Headquarters

HQ city

Cambridge

HQ country

United States

HQ region

North America

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations

Revenue model

1. LFM Open License (commercial license above threshold): Free LFM Open License v1.0 for commercial use by organizations under $10M annual revenue and for research/non-profit use; commercial license required above the $10M revenue threshold. Recurring licensing revenue from larger customers.
2. Enterprise custom solutions & multi-year strategic contracts: Multi-year strategic deals with large enterprises (e.g., Mercedes-Benz for in-car AI, Shopify for commerce search, Insilico Medicine for drug discovery) typically delivered as integrated custom solutions and ongoing support.
3. Liquid Apollo consumer mobile app: Direct-to-consumer mobile AI assistant for iOS and Android, supporting potential subscription or freemium monetization for end users.
4. Cloud and marketplace inference (Together AI, Modal, Bedrock, Hugging Face): Deployment of LFMs on third-party cloud platforms and marketplaces enables usage-based and managed-services revenue streams via partner infrastructure.
5. Managed services & LEAP platform access: Self-serve and managed access to the LEAP platform for building, deploying, and evaluating LFMs, providing ongoing platform-level recurring and managed-services revenue.

Pricing tiers

Go-to-market motion6 records

Distribution channels7 records

Marketing channels8 records

Core offering

Liquid AI develops and sells Liquid Foundation Models (LFMs), a family of efficient generative AI models (350M to 24B parameters) built on a proprietary liquid neural network architecture with Linear Input-Varying Systems and hybrid attention blocks. The company offers the LEAP developer platform for customizing and deploying these models on edge devices, laptops, and cloud, plus the Liquid Apollo consumer mobile app providing private on-device AI assistance. Models are available under the LFM Open License (free for organizations under $10M revenue and for research/non-profit use) and through enterprise contracts with major customers including Mercedes-Benz, Shopify, and Insilico Medicine.

Product overview

Liquid AI offers a unified edge-native AI platform anchored by its proprietary Liquid Foundation Models (LFMs) and supported by a deployment platform (LEAP), a consumer app (Liquid Apollo), and a research division (Liquid Labs). The LFMs form the core, spanning three generations (LFM1/LFM-7B, LFM2, and LFM2.5) with parameter counts from 350M to 24B and including specialized variants for vision-language (LFM2-VL, LFM2.5-VL-450M), audio (LFM2-Audio), retrieval (LFM2.5-ColBERT-350M, LFM2.5-Embedding-350M), and drug discovery (LFM2-2.6B-MMAI). LEAP provides the SDK and deployment infrastructure that lets enterprises and developers customize and ship LFMs to AMD Ryzen AI, Qualcomm, and other edge hardware with sub-100ms latency, while Liquid Apollo brings the same models directly to consumers via a mobile app. Industry solutions (e.g., Mercedes-Benz MBUX in-car assistant, Shopify commerce recommender, Insilico Medicine drug discovery) integrate LFMs into specific verticals. Together, these offerings position Liquid AI as a non-transformer alternative to frontier LLMs, optimized for efficient on-device deployment.

Differentiator

Problem solved

Functional benefit

Brands

• Liquid Foundation Models (LFM): Family of efficient, non-transformer generative AI foundation models for text, vision, audio, and other modalities, including LFM2 and LFM2.5 series.
• LEAP
• Liquid Apollo
• Liquid Nanos
• Liquid Labs

Products and services

• Liquid Foundation Models (LFMs) Family of generative AI foundation models using a proprietary liquid neural network architecture with Linear Input-Varying Systems (LIVs) and hybrid attention blocks, ranging from 350M to 24B parameters across LFM1, LFM2, and LFM2.5 generations, with multimodal variants for vision, audio, and retrieval. Designed for efficient on-device deployment with up to 1000x smaller footprint than frontier LLMs while delivering comparable quality.
• LEAP Developer platform for customizing and deploying Liquid Foundation Models on edge devices, laptops, and cloud. Supports deployment in approximately 10 lines of code, includes the LEAP SDK and workbench for fine-tuning and production deployment, and delivers sub-100ms TTFT on AMD Ryzen processors.
• Liquid Apollo Consumer mobile app providing private on-device AI assistance. Runs Liquid Foundation Models entirely on-device with no internet or API calls required. Supports 50+ languages for translation, summarization, idea generation, and on-the-go assistance. Available on iOS App Store and Google Play.
• Liquid Playground
• Liquid Labs Dedicated research team continuing the Liquid AI lineage of liquid neural network research originating from MIT CSAIL. Drives scientific breakthroughs enabling intelligent, personalized, and adaptive machines, with publications at ICLR, ICML, and NeurIPS.
• LFM2-2.6B-MMAI Scientific foundation model co-developed with Insilico Medicine for pharmaceutical research. A single 2.6B-parameter checkpoint trained to perform at state-of-the-art levels across multiple drug discovery subdomains (property prediction, molecular optimization, affinity prediction, retrosynthesis planning). Outperforms models 10x its size on 13 of 22 property prediction tasks and reaches 98.8% success rates on molecular optimization benchmarks.
• Mercedes-Benz MBUX Integration Embedded on-device AI deployment for Mercedes-Benz vehicles with third- and fourth-generation MBUX in North America. Integrates Liquid Foundation Models into the MB.OS software architecture to enhance the MBUX Virtual Assistant with advanced speech, voice control, and contextual reasoning without cloud dependency.
• Shopify Commerce Integration Multi-year Shopify partnership deploying sub-20ms LFMs for product search and generative recommender systems across Shopify's platform. Includes co-developed AI-powered search model outperforming larger systems in speed and effectiveness for product search and merchant experiences.
• Liquid Nanos Ultra-small foundation models (350M to 2.6B parameters) delivering GPT-4o-class performance on specific agentic tasks. Run directly on smartphones, laptops, and embedded devices without cloud dependency, matching the performance of models hundreds of times larger.
• LFM2.5 Family Current generation of Liquid Foundation Models pretrained on 28T tokens (up from 10T in LFM2). Includes Base, Instruct, Japanese, Vision-Language, and Audio-Language variants. Family members include LFM2.5-1.2B-Thinking, LFM2.5-350M, LFM2.5-8B-A1B, LFM2.5-VL-450M, LFM2.5-ColBERT-350M, and LFM2.5-Embedding-350M.
• LFM2 Family Second generation of Liquid Foundation Models introduced July 2025, described as the fastest on-device foundation models on the market. Includes LFM2-350M, LFM2-2.6B, LFM2-8B-A1B, LFM2-24B-A2B, LFM2-VL, LFM2-VL-3B, LFM2-Audio, LFM2-ColBERT-350M, and LFM2-2.6B-MMAI.

Quantifiable outcome

• Up to 1,000x smaller models than frontier LLMs while remaining capable
• +6 more outcomes

Named customers7 records

Segments7 records

Ideal customer profiles6 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration20 records

AI capability18 records

Feature8 records

Partnerships

28 partnerships are on record, tiered prospective, core, flagship, supporting and minor.

• AppleprospectiveStrategic or Co-development Partner · 1 May 2026Reportedly in discussions with Liquid AI to strengthen Apple's on-device AI capabilities, with potential acquisition under consideration as of May 2026 reporting. Not yet a confirmed commercial agreement.
• DPhi SpacecoreTechnology or Integration · 29 April 2026Technology/integration partnership in which Swiss startup DPhi Space successfully hosted Liquid AI's large language model on its Clustergate-2 orbital mission, demonstrating on-device AI in space environments.
• Mercedes-BenzflagshipOEM/ Whitelabel/ Licensing Partner · 23 April 2026Multi-year strategic partnership to embed Liquid AI's on-device foundation models into Mercedes-Benz vehicles, starting with MBUX. Production deployment targeted for H2 2026, with availability on 3rd and 4th gen MBUX vehicles back to the 2024 model year. Covers in-car voice assistant and intelligent features running natively on vehicle hardware.
• Insilico MedicinecoreStrategic or Co-development Partner · 3 March 2026Multi-year strategic partnership for AI-driven drug discovery. Liquid AI's LFM2-2.6B-MMAI model beats a 27B-parameter baseline on 13 of 22 molecular property prediction tasks and achieves 98.8% success in molecule optimization, supporting Insilico's generative chemistry and property prediction workloads.
• Together AIcoreTechnology or Integration · 25 February 2026Deployment partnership: Together AI has expanded its platform to support deployment of Liquid AI's LFM2-24B-A2B hybrid mixture-of-experts model, providing cloud inference access.
• AMDflagshipTechnology or Integration · 5 January 2026Technology partnership demonstrating on-device AI meeting summarization using AMD Ryzen hardware running Liquid AI LFMs; AMD also powers LEAP platform on AMD processors and showcases LFM2 models on AMD hardware.
• MITflagshipStrategic or Co-development Partner · 1 January 2026Ongoing research collaboration with MIT (including MIT CSAIL) on CompreSSM compression and core liquid neural network research; origin of the liquid neural network approach co-authored by Daniela Rus, Ramin Hasani, Mathias Lechner, and Alexander Amini.
• Max Planck InstitutesupportingStrategic or Co-development Partner · 1 January 2026Research collaboration with Max Planck Institute on the CompreSSM compression technique, supporting efficient state-space model design.
• ETH ZurichsupportingStrategic or Co-development Partner · 1 January 2026Research collaboration with ETH Zurich on CompreSSM compression, contributing to the academic underpinnings of Liquid's efficient foundation models.
• ShopifycoreStrategic or Co-development Partner · 13 November 2025Multi-year strategic partnership for commerce, deploying sub-20ms Liquid text models for on-device product search, merchant AI assistants, and catalog enrichment across the Shopify merchant ecosystem.
• Robotec.aisupportingTechnology or Integration · 23 October 2025Joint demonstration with AMD and Robotec.ai of LFM2-VL-3B in embedded autonomy, showcasing agentic robotics at the edge with Liquid AI's vision-language model.
• Brilliant LabscoreTechnology or Integration · 15 September 2025Technology integration partnership embedding Liquid AI's vision-language models into Brilliant Labs' Halo smart glasses, enabling on-device multimodal AI for wearable computing.
• Alef EducationcoreStrategic or Co-development Partner · 19 August 2025Strategic partnership to deploy Liquid AI models for personalized learning across 1.5M students and 14,000 schools in the MENA region, providing on-device educational AI at scale.
• G42coreStrategic or Co-development Partner · 17 June 2025Strategic partnership to deliver private, local, and efficient AI solutions to enterprises at scale across the UAE, leveraging Liquid AI's compact foundation models for regional enterprise and government-aligned workloads.
• QualcommsupportingTechnology or Integration · 1 January 2025Silicon partner integrating Liquid AI LFMs with Qualcomm platforms for on-device AI on mobile and edge hardware.
• Hugging FacesupportingChannel Partner/ Reseller/ Distributor · 1 January 2025Distribution and ecosystem platform hosting publicly released Liquid AI models (LFM2 family, MoE variants) for download and integration.
• Amazon BedrocksupportingChannel Partner/ Reseller/ Distributor · 1 January 2025Distribution through Amazon Bedrock marketplace, providing managed access to Liquid AI models on AWS infrastructure.
• ModalsupportingTechnology or Integration · 1 January 2025Cloud infrastructure partner providing serverless GPU runtime for deploying and serving Liquid AI models.
• OllamasupportingChannel Partner/ Reseller/ Distributor · 1 January 2025Edge runtime partner making Liquid AI models available via Ollama's local model runner ecosystem.
• LM StudiosupportingChannel Partner/ Reseller/ Distributor · 1 January 2025Local-inference runtime ecosystem partner supporting Liquid AI models on developer desktops and laptops.
• UnslothsupportingTechnology or Integration · 1 January 2025Fine-tuning ecosystem integration enabling efficient customization of Liquid AI models for developers.
• executorchsupportingTechnology or Integration · 1 January 2025Edge inference runtime (PyTorch-native) integration enabling deployment of Liquid AI models on mobile and edge devices.
• FastFlowLMsupportingTechnology or Integration · 1 January 2025Local AI inference runtime partner supporting Liquid AI models on consumer hardware for fast on-device inference.
• Nexa.aisupportingTechnology or Integration · 1 January 2025On-device AI ecosystem partner integrating Liquid AI models into the Nexa.ai edge inference stack.
• TXTsupportingStrategic or Co-development Partner · 1 January 2025Collaboration with TXT on integrating Liquid AI models into enterprise and industrial solutions.
• AILAminorStrategic or Co-development Partner · 1 January 2025Startup partner ecosystem collaboration supporting adoption of Liquid AI models by early-stage companies.
• CTCsupportingStrategic or Co-development Partner · 27 February 2024Collaboration with CTC on edge AI deployment of Liquid AI models for industrial and edge computing use cases.
• CapgeminisupportingImplementation/ SI/ Consulting Partner · 12 January 2024Collaboration with Capgemini to deploy enterprise AI solutions leveraging Liquid AI's compact foundation models for client engagements.

Scale indicators13 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Mistral AI: Paris-based foundation model developer offering open-weight and commercial LLMs (Mistral, Mixtral MoE) with strong developer adoption. Both companies compete in providing efficient, open-weight foundation models for enterprise and developer use cases.
• Sakana AI: Tokyo-based foundation model lab developing efficient AI models using novel non-transformer approaches inspired by nature. Strong architectural parallels to Liquid AI's liquid neural network lineage; both target efficient model training and deployment.
• AI21 Labs: Tel Aviv-based foundation model developer offering enterprise LLMs (Jamba, Jurassic) with emphasis on efficiency, long context, and structured outputs. Competes directly in the enterprise foundation model market with similar focus on deployable, production-ready models.
• Cohere: Enterprise-focused foundation model provider offering Command and Embed models with strong emphasis on retrieval, search, and private deployment. Direct overlap with Liquid's enterprise foundation model and retrieval (ColBERT/Embedding) product lines.
• Hugging Face: Open-source model hub and platform that hosts Liquid AI's models and competes as a distribution layer for foundation models. Comparable ecosystem play around open-weight model distribution, developer community, and inference offerings.

Broad incumbents

• Anthropic: Major foundation model developer behind Claude, serving enterprise and consumer with frontier LLMs. Competes in the enterprise AI market though with cloud-centric rather than edge-first deployment model.
• OpenAI: Frontier model developer behind GPT-4/GPT-5 family with massive enterprise and consumer distribution. Indirect competitor as enterprises choose between cloud frontier models and Liquid's on-device alternatives.
• Meta AI (Llama): Meta's open-weight Llama model family dominates the open-source LLM ecosystem. Sets the baseline against which Liquid's open-weight LFMs must differentiate on efficiency and on-device deployment.

Emerging players

• Aleph Alpha: European foundation model developer focused on sovereign, compliant, and explainable AI for enterprise and government. Comparable niche positioning around data sovereignty and European regulatory alignment, including G42-like enterprise/government use cases.
• Together AI: Cloud inference platform hosting open-weight models including Liquid's LFM2-24B-A2B. Partly partner, partly adjacent competitor in the open-model inference and fine-tuning ecosystem.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/Liquid4All/cookbook
• Xx.com/LiquidAI_
• LinkedInlinkedin.com/company/liquid-ai-inc

Compliance2 records

Revenue estimate

Valuation estimate

Number of profiles

Profiles6 records

Subsidiaries1 record

Funding overview

Funding rounds3 records

Investors12 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Liquid AI