Our partners at @LiquidAI released LFM3 on Tuesday. We tested the 1.7B text model on macOS inside Eney agent tasks, and it completed them with fewer tokens than a larger model. The interesting part is that for agents, a model that fits in about 1 GB changes where the work can happen. 🔥
kosovan retweeted
Most AI tools tell you HOW to get work done.
But you're still the one jumping between tabs, sorting files, and writing follow-ups.
Eney takes a different approach: an AI assistant for Mac designed to DO the tasks.
Here are 7 everyday workflows worth seeing. 🧵
kosovan retweeted
LFM3 is here! this is an insanely elaborate release, get yourself a coffee ☕️ spend a few hours going through the blog where we tell you everything about it.
The amount of useful intelligence you can pack inside 2B parameters is really insane!
the highlights for me:
> LFM3 architecture got out of our propriety automated hardware in the loop architecture search algorithm optimized for 14 different types of processors. This is exactly the vision we embarked on: building an intelligence fabric on the fragmented device world.
> LFM3-VL-2B is truly a beast performing on par with GPT 5.6 Luna and Gemini flash ultra! at 2B parameters, wtf!!
> It is sooo easy and chip to fine-tune LFM3 instances. you can now truly build specialized intelligence for anything.
> LFM3 is enjoyably fast to train, and deploy.
> LFM3 omni 2B packs all modalities in one model enabling reliable spacial intelligence on device!
> It is now the era of products. lets build together 💫🚀
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
Visa is letting agents execute approved supplier payments. The interesting part is “approved”. In Eney-to-Eney, which we are building, your Eney proposes and nothing leaves your Mac until you confirm. Therefore I think approval is part of the agent design, not a brake on it.
kosovan retweeted
The world must stop pretending russia is anything but a terrorist state.
1,600 guided aerial bombs dropped on Ukraine in a single week. Around 230 a day.
Why is russia allowed to get away with this? When did this become normal in the so-called civilized world?
One design decision in Eney-to-Eney I would like to argue about: if you decline a request from someone else’s Eney, the sender is never notified. The cost is they wait without an answer. The benefit is that saying no stays as cheap as ignoring an email. Any other thoughts?
kosovan retweeted
One-year-old Angelina, born in August 2025, died right in her stroller. Her mother, Inna, passed away in the hospital.
What were they guilty of? Why does the US President support this russian evil?
kosovan retweeted
Russians have just murdered a baby and two women that were walking in a public park in Kharkiv.
This is the face of genocide in the 21st century.
Now, under U.S. protection, Russia is inflicting massacres at the highest level of the war, all across Ukraine.
Mnemos is the memory layer. With the same conversations, answering model and judge for every system, it scored 89.0% on LongMemEval-S and led on 8 of 12 benchmarks. Memory is what lets an assistant be proactive instead of waiting for a prompt.
Under it, ELIX is our inference engine for Apple silicon. On identical model weights across 25 chip-and-model runs against six other engines, it was fastest in 19 and within 5% of the leader in 4 more. Every result is published, including the 2 where it trails.