LiquidAI Releases LFM2.5-VL-DSpark for Vision-Language Models
https://localmodelwatch.tsuchitsuchi.com/en/2026/09/25/liquidai-lfm2-5-vl-dspark-released/
https://localmodelwatch.tsuchitsuchi.com/en/2026/09/25/liquidai-lfm2-5-vl-dspark-released/
September 25, 2026 at 7:09 AM
LiquidAI Releases LFM2.5-VL-DSpark for Vision-Language Models
https://localmodelwatch.tsuchitsuchi.com/en/2026/09/25/liquidai-lfm2-5-vl-dspark-released/
https://localmodelwatch.tsuchitsuchi.com/en/2026/09/25/liquidai-lfm2-5-vl-dspark-released/
Liquid AI's LFM2.5-VL-450M, a vision-language model built for real-time reasoning on edge devices.
It processes a 512×512 image and returns structured outputs in ~240ms on-device.
- Blog: liquid.ai/blog/lfm2-5-...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai/login?callba...
It processes a 512×512 image and returns structured outputs in ~240ms on-device.
- Blog: liquid.ai/blog/lfm2-5-...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai/login?callba...
April 8, 2026 at 4:56 PM
Liquid AI's LFM2.5-VL-450M, a vision-language model built for real-time reasoning on edge devices.
It processes a 512×512 image and returns structured outputs in ~240ms on-device.
- Blog: liquid.ai/blog/lfm2-5-...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai/login?callba...
It processes a 512×512 image and returns structured outputs in ~240ms on-device.
- Blog: liquid.ai/blog/lfm2-5-...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai/login?callba...
Liquid will make a bigger one one day, I believe in them huggingface.co/LiquidAI/LFM...
LiquidAI/LFM2.5-VL-1.6B-Extract · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
September 16, 2026 at 5:59 PM
Liquid will make a bigger one one day, I believe in them huggingface.co/LiquidAI/LFM...
This shit is fast as fuck, doing some testing!
huggingface.co/LiquidAI/LFM...
Recent LFM releases have been pretty impressive imo.
huggingface.co/LiquidAI/LFM...
Recent LFM releases have been pretty impressive imo.
LiquidAI/LFM2.5-1.2B-Thinking · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
January 22, 2026 at 4:56 PM
This shit is fast as fuck, doing some testing!
huggingface.co/LiquidAI/LFM...
Recent LFM releases have been pretty impressive imo.
huggingface.co/LiquidAI/LFM...
Recent LFM releases have been pretty impressive imo.
good gods Liquid is not fucking around huggingface.co/LiquidAI/LFM...
this thing already runs at 100tps on an old consumer GPU and they're like. "nope too slow"
this thing already runs at 100tps on an old consumer GPU and they're like. "nope too slow"
LiquidAI/LFM2.5-2.6B-DSpark-GGUF · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
August 20, 2026 at 6:08 PM
good gods Liquid is not fucking around huggingface.co/LiquidAI/LFM...
this thing already runs at 100tps on an old consumer GPU and they're like. "nope too slow"
this thing already runs at 100tps on an old consumer GPU and they're like. "nope too slow"
LiquidAI released LFM2.5: 5 open-weight model instances from a single architecture:
> General-purpose instruct model
> Japanese-optimized chat model
> Vision-language model
> Native audio-language model (speech in/out)
> Base checkpoints for deep customization
> General-purpose instruct model
> Japanese-optimized chat model
> Vision-language model
> Native audio-language model (speech in/out)
> Base checkpoints for deep customization
January 6, 2026 at 7:42 AM
LiquidAI released LFM2.5: 5 open-weight model instances from a single architecture:
> General-purpose instruct model
> Japanese-optimized chat model
> Vision-language model
> Native audio-language model (speech in/out)
> Base checkpoints for deep customization
> General-purpose instruct model
> Japanese-optimized chat model
> Vision-language model
> Native audio-language model (speech in/out)
> Base checkpoints for deep customization
Liquid AI just made their vision-language model up to 3x faster with zero quality loss. A tiny 280M drafter accelerates the 3B VLM via speculative decoding, shipping day-one for SGLang, llama.cpp, and MLX. https://huggingface.co/blog/liquidai/lfm2-5-vl-dspark
September 25, 2026 at 6:05 AM
Liquid AI just made their vision-language model up to 3x faster with zero quality loss. A tiny 280M drafter accelerates the 3B VLM via speculative decoding, shipping day-one for SGLang, llama.cpp, and MLX. https://huggingface.co/blog/liquidai/lfm2-5-vl-dspark
📰 Hugging Face
Accelerating vision-language models with LFM2.5-VL-DSpark
https://huggingface.co/blog/LiquidAI/lfm2-5-vl-dspar#IA##AI##ML#ML
Accelerating vision-language models with LFM2.5-VL-DSpark
https://huggingface.co/blog/LiquidAI/lfm2-5-vl-dspar#IA##AI##ML#ML
September 24, 2026 at 3:00 PM
📰 Hugging Face
Accelerating vision-language models with LFM2.5-VL-DSpark
https://huggingface.co/blog/LiquidAI/lfm2-5-vl-dspar#IA##AI##ML#ML
Accelerating vision-language models with LFM2.5-VL-DSpark
https://huggingface.co/blog/LiquidAI/lfm2-5-vl-dspar#IA##AI##ML#ML
Latest open artifacts (#18): Arcee's 400B MoE, LiquidAI's underrated 1B model, new Kimi, and anticipation of a busy month
Tons of useful "niche" models and anticipation of big releases coming soon.
www.interconnects.ai/p/latest-ope...
Tons of useful "niche" models and anticipation of big releases coming soon.
www.interconnects.ai/p/latest-ope...
Latest open artifacts (#18): Arcee, LiquidAI and Moonshot ...
Tons of useful "niche" models and anticipation of big releases coming soon.
www.interconnects.ai
February 2, 2026 at 3:23 PM
Latest open artifacts (#18): Arcee's 400B MoE, LiquidAI's underrated 1B model, new Kimi, and anticipation of a busy month
Tons of useful "niche" models and anticipation of big releases coming soon.
www.interconnects.ai/p/latest-ope...
Tons of useful "niche" models and anticipation of big releases coming soon.
www.interconnects.ai/p/latest-ope...
> Blog: liquid.ai/blog/lfm2-5-...
> LFM2.5-2.6B: huggingface.co/LiquidAI/LFM...
> LFM2.5-2.6B-Base: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
> LFM2.5-2.6B: huggingface.co/LiquidAI/LFM...
> LFM2.5-2.6B-Base: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
LFM2.5-2.6B: Deploy Agents Everywhere — Blog
LFM2.5-2.6B is an on-device agentic model that plans, calls tools, and runs multi-step tasks at 220 tok/s in under 2.5 GB. Open weights on Hugging Face.
liquid.ai
August 4, 2026 at 2:29 PM
> Blog: liquid.ai/blog/lfm2-5-...
> LFM2.5-2.6B: huggingface.co/LiquidAI/LFM...
> LFM2.5-2.6B-Base: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
> LFM2.5-2.6B: huggingface.co/LiquidAI/LFM...
> LFM2.5-2.6B-Base: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
LiquidAI, MIT CSAIL offshoot, released Liquid Nanos as open-weight
> model size: 350M to 2.6B
> built on LFM2, v2 efficient model architecture
> perform competitively with models up to hundreds of times larger
> enable core agentic tasks: precise data extraction, multilingual translation, tool call
> model size: 350M to 2.6B
> built on LFM2, v2 efficient model architecture
> perform competitively with models up to hundreds of times larger
> enable core agentic tasks: precise data extraction, multilingual translation, tool call
September 25, 2025 at 2:02 PM
LiquidAI, MIT CSAIL offshoot, released Liquid Nanos as open-weight
> model size: 350M to 2.6B
> built on LFM2, v2 efficient model architecture
> perform competitively with models up to hundreds of times larger
> enable core agentic tasks: precise data extraction, multilingual translation, tool call
> model size: 350M to 2.6B
> built on LFM2, v2 efficient model architecture
> perform competitively with models up to hundreds of times larger
> enable core agentic tasks: precise data extraction, multilingual translation, tool call
V2: Hybrid architecture. LFM2.5 reasons in text, Piper speaks. Faster, more reliable. Still 100% local on a Pi5. No cloud required.
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
Kitsune Voice V2 Demo - 100% Local AI on Raspberry Pi 5 (LiquidAI LFM2)
YouTube video by delta : kitsune
www.youtube.com
January 15, 2026 at 1:49 PM
V2: Hybrid architecture. LFM2.5 reasons in text, Piper speaks. Faster, more reliable. Still 100% local on a Pi5. No cloud required.
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
LiquidAI (MIT SAIL offshoot) released the open-weights of LFM2-2.6B
- light-weight with 2.6B parameters
- fast, built by our v2 efficient architecture (short convs + group query attention)
- Trained on 10T tokens32k context length
- open-weight, multilingual, great instruction following
- light-weight with 2.6B parameters
- fast, built by our v2 efficient architecture (short convs + group query attention)
- Trained on 10T tokens32k context length
- open-weight, multilingual, great instruction following
September 23, 2025 at 8:54 PM
LiquidAI (MIT SAIL offshoot) released the open-weights of LFM2-2.6B
- light-weight with 2.6B parameters
- fast, built by our v2 efficient architecture (short convs + group query attention)
- Trained on 10T tokens32k context length
- open-weight, multilingual, great instruction following
- light-weight with 2.6B parameters
- fast, built by our v2 efficient architecture (short convs + group query attention)
- Trained on 10T tokens32k context length
- open-weight, multilingual, great instruction following
LiquidAI cooked with LFM2-VL. At the risk of sounding like an X AI influencer, don't sleep on this model. I'm finetuning right now on Catmus. A small test over night on only 3k examples is showing remarkable improvement. Training now on 150k samples. I see this as potentially replacing TrOCR.
August 13, 2025 at 3:01 PM
LiquidAI cooked with LFM2-VL. At the risk of sounding like an X AI influencer, don't sleep on this model. I'm finetuning right now on Catmus. A small test over night on only 3k examples is showing remarkable improvement. Training now on 150k samples. I see this as potentially replacing TrOCR.
- Blog: liquid.ai/blog/lfm2-24...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai
February 25, 2026 at 4:01 PM
- Blog: liquid.ai/blog/lfm2-24...
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai
- Model: huggingface.co/LiquidAI/LFM...
- Demo: playground.liquid.ai
- Blog: liquid.ai/blog/lfm2-5-...
- Weights: huggingface.co/LiquidAI/LFM...
- Docs: docs.liquid.ai
- Playground: playground.liquid.ai
- Weights: huggingface.co/LiquidAI/LFM...
- Docs: docs.liquid.ai
- Playground: playground.liquid.ai
LFM2.5-8B-A1B: an Even Better on-Device Mixture-of-Experts | Liquid AI
Today, we’re releasing LFM2.5-8B-A1B, a high-throughput edge model optimized for fast, reliable tool calling and complex instruction following on consumer hardware, delivering compressed performance c...
liquid.ai
May 30, 2026 at 5:10 PM
- Blog: liquid.ai/blog/lfm2-5-...
- Weights: huggingface.co/LiquidAI/LFM...
- Docs: docs.liquid.ai
- Playground: playground.liquid.ai
- Weights: huggingface.co/LiquidAI/LFM...
- Docs: docs.liquid.ai
- Playground: playground.liquid.ai
> Blog post: liquid.ai/blog/lfm2-5-...
> LFM2.5-230M: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
> LFM2.5-230M: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
LFM2.5-230M: Built to Run Anywhere | Liquid AI
Meet LFM2.5-230M, Liquid AI’s smallest model yet: a fast, open-weight foundation model for fine-tuning, edge deployment, tool use, and data extraction.
liquid.ai
June 25, 2026 at 6:11 PM
> Blog post: liquid.ai/blog/lfm2-5-...
> LFM2.5-230M: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
> LFM2.5-230M: huggingface.co/LiquidAI/LFM...
> Docs: docs.liquid.ai
Liquid LFM model is one of those good reasoning and coding models. Worth exploring for the free AI model.
#buildinpublic #aicodingtools #tutorial #vibecoding #ai
#buildinpublic #aicodingtools #tutorial #vibecoding #ai
LiquidAI: LFM2.5-1.2B - Free AI Coding Model
YouTube video by Helpfolder
www.youtube.com
July 20, 2026 at 9:32 AM
Liquid LFM model is one of those good reasoning and coding models. Worth exploring for the free AI model.
#buildinpublic #aicodingtools #tutorial #vibecoding #ai
#buildinpublic #aicodingtools #tutorial #vibecoding #ai
from the other direction, LFM models often punch above their weight, and I've seen 2.5-1.2B-thinking outsmart (ok, answer a deliberately misleading question correctly) gemma e4b. huggingface.co/collections/... (they have a 8b-a1b too if moe is ok)
💧 LFM2.5 - a LiquidAI Collection
Collection of post-trained and base LFM2.5 models.
huggingface.co
August 15, 2026 at 4:11 PM
from the other direction, LFM models often punch above their weight, and I've seen 2.5-1.2B-thinking outsmart (ok, answer a deliberately misleading question correctly) gemma e4b. huggingface.co/collections/... (they have a 8b-a1b too if moe is ok)
Working examples of a smarter Google Home. On a Pi. Repo in comments of the post - fork it! The audio variant LLM is a cool concept.
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
V1: Pure audio-to-audio using LiquidAI's LFM2-Audio. Single model handles everything. Pi5, no cloud, $150 hardware. Experimental but it works. Repo in comments.
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
www.youtube.com/shorts/kNyeg...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
www.youtube.com/shorts/kNyeg...
Kitsune Voice V1 Demo - 100% Local AI on Raspberry Pi 5 -LFM2-Audio Omni-Modal (Audio-to-Audio)
YouTube video by delta : kitsune
www.youtube.com
January 15, 2026 at 1:54 PM
Working examples of a smarter Google Home. On a Pi. Repo in comments of the post - fork it! The audio variant LLM is a cool concept.
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
February 10, 2026 at 8:47 PM
Working examples of a smarter Google Home. On a Pi. Repo in comments of the post - fork it!
This version is smoother. And this is early days with the LFM2 series 🦊
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
This version is smoother. And this is early days with the LFM2 series 🦊
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
V2: Hybrid architecture. LFM2.5 reasons in text, Piper speaks. Faster, more reliable. Still 100% local on a Pi5. No cloud required.
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
Repo in comments.
www.youtube.com/shorts/4fMue...
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
Kitsune Voice V2 Demo - 100% Local AI on Raspberry Pi 5 (LiquidAI LFM2)
YouTube video by delta : kitsune
www.youtube.com
January 15, 2026 at 1:56 PM
Working examples of a smarter Google Home. On a Pi. Repo in comments of the post - fork it!
This version is smoother. And this is early days with the LFM2 series 🦊
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst
This version is smoother. And this is early days with the LFM2 series 🦊
#BuildInPublic #LocalAI #OpenSource #RaspberryPi #EdgeComputing #LiquidAI #AIAccessibility #LocalFirst