Liquid AI (@liquidai) / X

Liquid AI

685 posts

Liquid AI

@liquidai

Build efficient general-purpose AI at every scale.

Cambridge, MA

Joined March 2023

Pinned
Liquid AI
@liquidai
May 28
Today, we're releasing LFM2.5-8B-A1B, a device-optimized model designed to power real-life applications on phones, laptops, PCs, robots, and fast & lightweight server-side use-cases. > 8B MoE, 1.5B active > Expanded 128K context > LFM2.5 flagship hybrid MoE architecture >
1.3M
Liquid AI reposted
Alexander Amini
@xanamini
Jun 25
🐘 @liquidai
Tim Seyde
@timseyde
Jun 25
Dumbo's first steps — LFM2.5-230M doing multi-step tool-calling over pre-trained skills provided by @nvidia SONIC. Same small model, many different use cases.
2.8K
Liquid AI
@liquidai
Jun 26
A human genome is billions of base pairs. You can't use one as model context unless the model is efficient enough to handle it. Our CTO Mathias Lechner, @mlech26l, sits down with co-founder and Chief Science Officer Alexander Amini, @xanamini, on what it takes to build foundation
00:00
6.3K
Liquid AI reposted
sabesh 📟
@sabeshbharathi
Jun 26
LFM launched their ultra small, somewhat capable tool-use model. And with MLX - the speed on this is INCREDIBLE. Trying to see what is an apt agentic use case for this one.
00:00
Liquid AI
@liquidai
Jun 25
Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic tasks on phones, robots, home and network automation devices. > 230M parameters, built on the LFM2 architecture > Pre-trained on 19T tokens, with a 32K context
2.9K
Liquid AI reposted
Derya Unutmaz, MD
@DeryaTR_
Jun 26
Liquid AI does it again! They just released a 230M model LFM2.5 that can run anywhere for agentic tasks, maybe even on a Raspberry Pi!☺️
Liquid AI
@liquidai
Jun 25
Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic tasks on phones, robots, home and network automation devices. > 230M parameters, built on the LFM2 architecture > Pre-trained on 19T tokens, with a 32K context
19K
Liquid AI reposted
Xenova
@xenovacom
Jun 25
While we eagerly await Fable 5's return, our agentic WebGPU kernel optimization framework kept running. Opus 4.8 picked up where Fable left off, pushing Liquid AI's new LFM2.5 230M to an unbelievable 1,400 tok/s... running locally in your browser. Don't blink or you'll miss it.
00:00
164K
Liquid AI reposted
Ramin
@ramin_m_h
Jun 25
1400 tok/s on your browser 🤣 🐐 ed as always @xenovacom!
Xenova
@xenovacom
Jun 25
While we eagerly await Fable 5's return, our agentic WebGPU kernel optimization framework kept running. Opus 4.8 picked up where Fable left off, pushing Liquid AI's new LFM2.5 230M to an unbelievable 1,400 tok/s... running locally in your browser. Don't blink or you'll miss it.
00:00
2.8K
Liquid AI reposted
Sarah Mulligan
@sarahintheloop
Jun 25
Releasing a groundbreakingly small model. Adding it to a humanoid robot, and naming that robot Dumbo. That's just a normal day's work at @liquidai , and it's incredible to be working at a place that moves so quickly with its innovations. Read more about the debut of LFM2.5-230M,
00:00
1.8K
Liquid AI reposted
David Hendrickson
@TeksEdge
Jun 25
🧠Localmaxxers 👀 LLM built for CPU performing at 200+ tps! I have to test this on a Raspberry Pi. Liquid AI dropped LFM2.5-230M, a 230M model built from the ground up for CPU and edge devices, not GPUs. ⚡ 213 tokens/s on Galaxy S25 Ultra (CPU) ⚡ 42 tokens/s on Raspberry Pi
Liquid AI
@liquidai
Jun 25
Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic tasks on phones, robots, home and network automation devices. > 230M parameters, built on the LFM2 architecture > Pre-trained on 19T tokens, with a 32K context
7.5K
Liquid AI reposted
Tim Seyde
@timseyde
Jun 25
Dumbo's first steps — LFM2.5-230M doing multi-step tool-calling over pre-trained skills provided by @nvidia SONIC. Same small model, many different use cases.
Liquid AI
@liquidai
Jun 25
Replying to @liquidai
As an early look at ongoing work, we deployed LFM2.5-230M on a Unitree G1, running entirely on-device on its onboard @nvidia Jetson Orin. The model acts as a skill-selection layer, taking in natural-language instructions and decomposing them into sequences of tool calls. After
00:00
5.5K
Liquid AI reposted
Jimmy Smith
@jimmysmith1919
Jun 25
We are releasing the smallest member of the LFM2 family yet: LFM2.5-230M. It is meant for specialized, extremely low latency tasks. Check out the demo of it being used as a skill-selection layer, turning prompts into tool calls on a Unitree G1:
00:00
Liquid AI
@liquidai
Jun 25
Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic tasks on phones, robots, home and network automation devices. > 230M parameters, built on the LFM2 architecture > Pre-trained on 19T tokens, with a 32K context
5.6K
Liquid AI reposted
Sergio Paniego
@SergioPaniego
Jun 25
you can now train @liquidai's LFM2-VL in TRL GRPO and RLOO included, with an example script
14K
Liquid AI reposted
LMSYS Org
@lmsysorg
Jun 25
🎉 Meet LFM2.5-230M from @liquidai, their smallest model yet at 230M params, but it punches way above its weight. Day 0 Support is live on SGLang! Built on the LFM2 architecture for on-device deployment: > Blazing-fast inference, runs everywhere, from cloud GPUs to low-cost CPUs
Liquid AI
@liquidai
Jun 25
Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic tasks on phones, robots, home and network automation devices. > 230M parameters, built on the LFM2 architecture > Pre-trained on 19T tokens, with a 32K context
4.7K