World & Business

Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

Liquid AI unveils six LFM3 models spanning text, vision, audio and decision tasks · Digg

Announcement

Liquid AI unveils six LFM3 models spanning text, vision, audio and decision tasks

Liquid AI pitches a tri-hybrid design, single-GPU fine-tuning and on-device use without detailing hardware requirements.

TLDR

Liquid AI’s Oct. 27 LFM3 announcement describes six models covering text, vision, audio, combined modalities and decision tasks. The company lists a 256K context window for the text model, video input for its vision-language model, and voice chat and reasoning in 18 languages for its audio model. It also pitches on-device use and single-GPU fine-tuning, but the announcement does not provide detailed hardware requirements or independent performance results.

Combined views

25.3K

14 Sources, first seen 2h ago

Combined views

25.3K

14 Sources, first seen 2h ago

429 likes

Useful links

huggingface

Build specialized multimodal models with LFM3

Liquid AI · YouTube

LFM3-VL-2B for Robotics: Perception, Navigation and Manipulation

Liquid Docs

LFM3-VL-2B - Liquid Docs

Alex Volkov from ThursdAI · YouTube

Maxime Labonne interview: decision models with Liquid AI

Liquid Docs

Audio Models - Liquid Docs

Liquid Docs

LFM3-Omni-2B - Liquid Docs

Liquid Docs

d1-omni-2B - Liquid Docs

2h ago

first seen 2h ago

429 likes

24 comments

214 saves

176 reposts

24 comments

214 saves

176 reposts

Related

2 weeks earlier

Liquid AI releases two open-weight multimodal decision models

Open d1 pairs a 3B text-and-vision model with a 600M model for image or audio input.

Liquid AI announced LFM3 on Oct. 27, introducing six models for text, vision, audio, combined modalities and decision tasks. The company is pitching a family that users can customize and run on-device, rather than a single model for every kind of input.

Featured Source

Related

2 weeks earlier

Liquid AI releases two open-weight multimodal decision models

Open d1 pairs a 3B text-and-vision model with a 600M model for image or audio input.

Sentiment

What Sources said

@liquidai (X) presents LFM3 as suited to specialized intelligence; @rohanpaul_ai (X) credits its tri-hybrid design for efficiency. @omarsar0 (X) sees promise for always-on local agents, while @Plinz (X) considers the models compact and suited to smaller devices.

@liquidai

3 weeks earlier

Liquid AI unveils d1, claimed to be the first model to outperform Jev on Hugging Face's Decision Index

Liquid AI says d1 wins on multilingual evaluations, is more robust against prompt injection and handles longer inputs more effectively. It lists access through the Liquid API.

3 weeks earlier

Liquid AI unveils d1, claimed to be the first model to outperform Jev on Hugging Face's Decision Index

Liquid AI says d1 wins on multilingual evaluations, is more robust against prompt injection and handles longer inputs more effectively. It lists access through the Liquid API.

· X

Positive

@grok · XPositive

@qiuqiu23126 · XPositive

@rohanpaul_ai · XPositive

@Plinz · XPositive

@omarsar0 · XPositive

Useful Links

huggingface

Build specialized multimodal models with LFM3

Liquid Docs

LFM3-VL-2B - Liquid Docs

Liquid Docs

Audio Models - Liquid Docs

Liquid Docs

LFM3-Omni-2B - Liquid Docs

Liquid Docs

d1-omni-2B - Liquid Docs

Related Videos

LFM3-VL-2B for Robotics: Perception, Navigation and ManipulationLiquid AI · YouTube
Maxime Labonne interview: decision models with Liquid AIAlex Volkov from ThursdAI · YouTube

Today's Rank

#5

Today's Rank

#5

According to Liquid AI, the LFM3-1.7B text model has a 256K context window. LFM3-VL-2B is a vision-language model with video-input support, while LFM3-Audio-2B supports tool calls, voice chat and reasoning in 18 languages. LFM3-Omni-2B combines vision, audio and language.

The remaining models, d1-2B and d1-Omni-2B, are designed for decision tasks. Liquid AI says they produce typed outputs and calibrated probabilities in a single pass; the Omni version also takes audio input. The company describes the family as using a tri-hybrid architecture optimized for 14 processor classes across GPUs, CPUs and NPUs, and says it can be fine-tuned on one GPU.

Liquid AI also claims its vision-language model performs competitively with larger and prominent models. Its announcement does not set out evaluation methods, independent performance results or detailed hardware requirements for the on-device and fine-tuning claims.

14 Sources

@atomic_chat_hqRun LFM3-1.7B locally on 8GB Mac 💻 We released 4 quants of @liquidai's new 1.7B agent model on Hugging Face from 1.8GB down to 1.1GB All of them fit on an 8GB Mac, so take the 1.8GB Q8_0 for the closest match to the original GGUFs 👇 https://huggingface.co/AtomicChat/LFM3-1.7B-GGUF2h

@_akhaliqRT @liquidai: Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encod…2h

@lmstudioLFM3 is live in LM Studio! Fits in about 1 GB of memory. Model page: https://lmstudio.ai/models/lfm32h

@huggingfaceRT @liquidai: Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encod…2h

@omarsar0You don't see releases like this every day. Liquid AI just released LFM3, a family of ~2B open-weight models for text, vision, audio, omni, and decision tasks, all on one shared backbone. LFM3-1.7B is trained to run in agent harnesses like Hermes Agent and Pi, with a 256K context. On WebArena-Verified, it scores 38.4 vs. 28.9 for Qwen3.5-9B in Liquid's evaluation. Models can be fine-tuned on a single GPU. This series of models has huge potential for always-on local agents.2h

@rohanpaul_aiLiquid AI just launched LFM3, an open-weight model family whose 1.7B text model beats its 2.6B predecessor. The gain comes from a tri-hybrid design: of 24 layers, 13 are short convolutions, 5 use 256-token sliding-window attention and 6 attend globally. That efficiency carries into agent benchmarks run by Liquid, where the 1.7B model scores 41.29 on AutomationBench while Qwen3.5-9B manages 15.10.1h

@adsimple🧪 Cognition testet Coding-Modelle mit offenen Aufgaben 📱 Liquid AI stellt LFM3 für lokale KI vor 🌍 Google zeigt Genie 4 Das KI-Abendupdate von AdSimple, gesponsert von ChatReact. https://x.com/i/article/21151280696775352331h

@qiuqiu23126兄弟们，卧槽！Liquid AI今天直接甩出LFM3家族，6款开源2B级模型共用同一骨干，文本模型才1.73B参数，却能跑256K上下文、支持16种语言，Hugging Face上就能用。小模型塞进手机、笔记本也不费劲，这波是真想把AI塞进日常设备里。你们最想先拿它干点啥？ #LiquidAI #LFM3 #OpenSource1h

@pricepertokenHugging Face published a guide on building specialized multimodal models with LFM3.1h

@PlinzLiquid is releases a suite of amazing, compact models to use on smaller devices1h

Sentiment

What Sources said

@liquidai (X) presents LFM3 as suited to specialized intelligence; @rohanpaul_ai (X) credits its tri-hybrid design for efficiency. @omarsar0 (X) sees promise for always-on local agents, while @Plinz (X) considers the models compact and suited to smaller devices.

@liquidai · XPositive

Useful Links

huggingface

Build specialized multimodal models with LFM3

Related Videos

@grok · XPositive

@qiuqiu23126 · XPositive

@rohanpaul_ai · XPositive

@Plinz · XPositive

@omarsar0 · XPositive

Liquid Docs

LFM3-VL-2B - Liquid Docs

Liquid Docs

Audio Models - Liquid Docs

Liquid Docs

LFM3-Omni-2B - Liquid Docs

Liquid Docs

d1-omni-2B - Liquid Docs

LFM3-VL-2B for Robotics: Perception, Navigation and ManipulationLiquid AI · YouTube

Maxime Labonne interview: decision models with Liquid AIAlex Volkov from ThursdAI · YouTube

14 Sources

@atomic_chat_hqRun LFM3-1.7B locally on 8GB Mac 💻 We released 4 quants of @liquidai's new 1.7B agent model on Hugging Face from 1.8GB down to 1.1GB All of them fit on an 8GB Mac, so take the 1.8GB Q8_0 for the closest match to the original GGUFs 👇 https://huggingface.co/AtomicChat/LFM3-1.7B-GGUF2h

@_akhaliqRT @liquidai: Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encod…2h

@lmstudioLFM3 is live in LM Studio! Fits in about 1 GB of memory. Model page: https://lmstudio.ai/models/lfm32h

@huggingfaceRT @liquidai: Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encod…2h

@omarsar0You don't see releases like this every day. Liquid AI just released LFM3, a family of ~2B open-weight models for text, vision, audio, omni, and decision tasks, all on one shared backbone. LFM3-1.7B is trained to run in agent harnesses like Hermes Agent and Pi, with a 256K context. On WebArena-Verified, it scores 38.4 vs. 28.9 for Qwen3.5-9B in Liquid's evaluation. Models can be fine-tuned on a single GPU. This series of models has huge potential for always-on local agents.2h

@rohanpaul_aiLiquid AI just launched LFM3, an open-weight model family whose 1.7B text model beats its 2.6B predecessor. The gain comes from a tri-hybrid design: of 24 layers, 13 are short convolutions, 5 use 256-token sliding-window attention and 6 attend globally. That efficiency carries into agent benchmarks run by Liquid, where the 1.7B model scores 41.29 on AutomationBench while Qwen3.5-9B manages 15.10.1h

@adsimple🧪 Cognition testet Coding-Modelle mit offenen Aufgaben 📱 Liquid AI stellt LFM3 für lokale KI vor 🌍 Google zeigt Genie 4 Das KI-Abendupdate von AdSimple, gesponsert von ChatReact. https://x.com/i/article/21151280696775352331h

@qiuqiu23126兄弟们，卧槽！Liquid AI今天直接甩出LFM3家族，6款开源2B级模型共用同一骨干，文本模型才1.73B参数，却能跑256K上下文、支持16种语言，Hugging Face上就能用。小模型塞进手机、笔记本也不费劲，这波是真想把AI塞进日常设备里。你们最想先拿它干点啥？ #LiquidAI #LFM3 #OpenSource1h

@pricepertokenHugging Face published a guide on building specialized multimodal models with LFM3.1h

@PlinzLiquid is releases a suite of amazing, compact models to use on smaller devices1h