It all began in December 2023 when Liquid AI, an MIT spin-off co-founded by a quartet of MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) scientists, raised $46.6 million – a substantial amount for a seed round – for the development of liquid neural networks, much smaller but potentially no less capable AI models that require far less computing power to run.
A year later, in December 2024, the AI startup offering a fundamentally different AI architecture raised $250 million in an early-stage funding round led by Advanced Micro Devices (AMD).
With more investments coming in, will Liquid deliver on its mission “to build the most capable and efficient AI system at every scale” in 2025? Let’s find out.
Key Takeaways
- Liquid AI focuses on adaptable liquid neural networks that require less computational power than traditional AI.
- The company raised $297 million in funding, including $250 million led by AMD, and is valued at $2.3 billion.
- Liquid foundation models (LFMs), like the newly launched LFM-7B, deliver advanced performance with minimal memory use and real-time adaptability.
- LFM-7B supports enterprise chat, coding, and instruction following while excelling in energy-efficient on-device AI applications.
- Liquid AI emphasizes accessibility for businesses of all sizes, with models available on platforms like AWS Marketplace and Liquid Playground.
What Is Liquid AI?
Liquid AI, an MIT spin-off, develops generative AI models based on a fundamentally different architecture than traditional transformer-based AI – GPTs.
The startup has four co-founders: Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela Rus. They all have outstanding backgrounds in AI research and machine learning (ML).
Liquid AI Co-Founders
- Daniela Rus: Co-founder & Director of Liquid AI, Professor of Electrical Engineering and Computer Science, and Director of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. Rus’s research interests are in robotics, mobile computing, and data science.
- Ramin Hasani: Co-founder & CEO of Liquid AI and a machine learning Research Affiliate at CSAIL. Ramin’s research focuses on robust deep learning and decision-making in complex dynamical systems.
- Mathias Lechner: Co-founder & CTO of Liquid AI and Research Affiliate at CSAIL. Lechner focuses on developing robust and trustworthy machine-learning models.
- Alexander Amini: Co-founder & Chief Science Officer of Liquid AI and Postdoctoral Researcher in the Computer Science and Artificial Intelligence Laboratory (CSAIL). Amini leads research to develop the science and engineering of autonomy and its applications to safe decision-making for autonomous agents.
Groundbreaking Research Behind Liquid AI: The Beginning of New AI?
What Is a Liquid Neural Network?
A liquid neural network is a class of brain-inspired machine learning systems designed to remain adaptable and robust even after initial training. According to Liquid AI, they are:
“Universal approximators, expressive continuous-time machine learning systems for sequential data, parameter efficient in learning new skills, causal and interpretable, and when linearized, they can efficiently model very long-term dependencies in sequential data.”
But how exactly do liquid neural networks differ from traditional AI models?
Key features of liquid neural networks:
- Dynamic architecture: Liquid neural networks have a flexible, “fluid” structure, which adapts as they process new data.
- Real-time learning: They can adjust their parameters in real time, unlike traditional AI models, which have fixed weights and connections once trained.
- Adaptability: Liquid networks are better for handling changing situations and time-series data.
- Interpretability: Liquid networks are smaller and simpler, making them easier to understand than traditional models.
- Computational efficiency: Liquid networks can work with fewer parameters, requiring fewer computational resources.
Simply put, the fewer parameters, the less computing power is needed to train and run the model. This could potentially make liquid neural network architecture attractive for further AI development.
From LNN to LFM: What Is a Liquid Foundation Model?
Moving further, the team introduced liquid foundation models (LFMs), a new generation of generative AI models that achieve advanced performance while maintaining a smaller memory footprint and more efficient inference.
On January 20, 2025, released a new model – LFM-7B – the “best-in-class language model in English, Arabic, and Japanese” designed to be the “substrate for private enterprise chat, code, fast instruction following, and agentic workflows.”
Highlighting the 7B’s key capabilities, Liquid AI mentioned the following:
- LFM-B7, maintaining expansive knowledge and reasoning, is optimized for private enterprise chat, coding, fast instruction following, and agentic workflows.
- LFM-7B has a minimal memory footprint compared to other architectures.
- It outperforms other models in its size class.
Speaking of memory efficiency, the company highlighted several key features, including long-context understanding, energy-efficient inference, and high-throughput deployments on local devices. They said:
“Consequently, LFM-7B significantly increases value for end users in applications such as private enterprise chat, secure code generation, fast instruction following, long document analysis, energy-efficient on-device AI assistants, and multi-step agentic workflows.”
STAR Architecture
Another substantial part of Liquid AI’s innovation worth a special mention is a Scalable Transformer Alternative Representations (STAR) architecture, which aims to boost AI efficiency and scalability.
According to Liquid AI, a new STAR model architecture outshines Transformers. For reference, Transformer is the technology behind most of the current generative AI models, introduced by Google researchers in 2017.
Excited to unveil STAR: Synthesis of Tailored Architectures! 🌟
We develop an evolutionary algorithm to automate neural architecture design, optimizing for quality, size, and latency — tailored to desired metrics & hardware specs.
Synthesized models outperform Transformers and… pic.twitter.com/k5A8wgiN08
— Alexander Amini (@xanamini) December 2, 2024
Liquid AI Stock: Can You Invest in Liquid AI?
No, Liquid AI is a private company with no stock publicly available for trading.
Despite some circulating rumors about a potential IPO, the company has not made any official announcements about plans to go public so far.
Liquid AI Funding: $297 Million
Liquid AI has raised a total funding of $297 million over two rounds.
The latest $250 million Series A funding round, led by AMD – a leader in the AI GPU market – will enable the company to fasten the deployment of its models across various real-life scenarios in telecommunications, financial services, e-commerce, and biotechnology.
Liquid AI (@LiquidAI_) has raised $250 million in a Series A round led by AMD (@AMD) at a $2.3 billion valuation. Liquid AI (@LiquidAI_), Cambridge, Massachusetts, United States, was founded in 2023 by Alexander Amini (@xanamini), Daniela Rus, Mathias Lechner (@mlech26l), and… pic.twitter.com/05f9iDQoNR
— Silicon Valley Investclub (@Investclubsv) December 16, 2024
A year earlier, the startup secured $46.6 million in a seed funding round.
Liquid AI Valuation: $2.3 Billion
Liquid AI gained prominence as one of the most innovative AI research startups. Its latest valuation of around $2.3 billion propelled it to the top AI unicorns.
Where’s Next?
Looking ahead, Liqud AI has robust plans. The latest funding round will enable the team to expand its research, engineering, and operational capabilities and focus on making its AI models accessible to businesses of all sizes.
Ramin Hasani, CEO and co-founder of Liquid AI said at an exclusive MIT Event:
“Our Liquid Foundation Models elevates the scaling laws for general-purpose AI systems at every scale for any data modality. Our first series of language LFMs achieve state-of-the-art performance at every scale while maintaining a small on-device memory footprint. This opens new possibilities for real-time, local AI applications, allowing our enterprise customers to harness AI without the limitations of heavy cloud dependence or extensive memory requirements.”
Today, users can already try out the company’s liquid foundation models (LFMs) on AWS Marketplace.
Starting today, you can try out our Liquid Foundation models (LFMs) on @awscloud AWS Marketplace, launched at #reinvent2024
If you like them, get in touch with us!
We are excited to hearing your feedback on LFMs! Also get ready for a couple of small and big product updates in… pic.twitter.com/RvTsYR0nYe
— Liquid AI (@LiquidAI_) December 4, 2024
LFMs are also available on Liquid Playground, Lambda (Chat UI and API), Perplexity Labs, and Cerebras Inference. According to the company’s website, the LFM stack is optimized for NVIDIA, AMD, Qualcomm, Cerebras, and Apple hardware.
The Bottom Line
Closing up with the words of Daniela Rus, Liquid AI’s Director: “Today’s AI has a ceiling…Let’s not settle for the current offering.”
MIT CSAIL Director Daniela Rus @TEDTalks: How AI will step off the screen and into the real world. https://t.co/NkuODF5MNf pic.twitter.com/KAZ6u8o8z9
— MIT CSAIL (@MIT_CSAIL) April 19, 2024
It’s yet to be seen whether Liquid AI will deliver on its promise to “deploy frontier-AI-powered solutions that are available to everyone” and seamlessly integrate AI at all enterprises. For now, they are moving forward progressively.
FAQs
Who is the CEO of Liquid AI?
How much is Liquid AI worth?
What is the latest Liquid AI model?
References
- Liquid AI: A New Generation of AI Models from First Principles (Liquid)
- We raised $250M to scale capable and efficient general-purpose AI (Liquid)
- From Liquid Neural Networks to Liquid Foundation Models (Liquid)
- Liquid Foundation Models: Our First Series of Generative AI Models (Liquid)
- Liquid AI on X (X)
- Liquid AI’s new STAR model architecture outshines Transformer efficiency | Liquid AI (Liquid)
- Liquid AI to Unveil First Products Built on Liquid Foundation Models (LFMs) at Exclusive MIT Event | Liquid AI (Liquid)