what is the fastest decision model you say?
9 7 1 78 6.6K 22
Liquid AI
@liquidai
Build efficient general-purpose AI at every scale.
38.6KFollowers
64Following
1.2KPosts
221Media
2.8KLikes
@liquidai reposted
@liquidai reposted
now available as a HF space 🤗
i challenge @songdng and @SinoueG to beat my score.
(brb getting a new wardrobe)
huggingface.co/spaces/LiquidA…
i challenge @songdng and @SinoueG to beat my score.
(brb getting a new wardrobe)
huggingface.co/spaces/LiquidA…
clem 🤗 1 5 1 18 3K 0
@liquidai reposted
leonie@helloiamleonie · 13hnow available as a HF space 🤗
i challenge @songdng and @SinoueG to beat my score.
(brb getting a new wardrobe)
huggingface.co/spaces/LiquidA…
i challenge @songdng and @SinoueG to beat my score.
(brb getting a new wardrobe)
huggingface.co/spaces/LiquidA…
Open quoted post →
3 2 1 20 2.5K 1
Vercel Developers@vercel_dev · 7hd1 from @liquidai is live on AI Gateway.
𝚕𝚒𝚚𝚞𝚒𝚍/𝚍𝟷
Make decisions with vision support to classify, route, and score.
vercel.com/changelog/liqu…
Open quoted post →𝚕𝚒𝚚𝚞𝚒𝚍/𝚍𝟷
Make decisions with vision support to classify, route, and score.
vercel.com/changelog/liqu…
2 4 0 44 2.3K 2
@liquidai reposted
Wow. Less than a month ago, Jev released and decision models took over X. 3 weeks later, @liquidai shipped a 3B that goes head to head with Jev -- on a Mac, 30 ms latency, and *with* vision...
The frontier is reaching local hardware faster than anyone could've anticipated.
The frontier is reaching local hardware faster than anyone could've anticipated.
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
3 6 0 32 1.8K 11
@liquidai reposted
1:22 3 1 1 5 1.1K 0
@liquidai reposted
Good conversation from our COO Jeffrey Li on theCUBE at Fully Connected. The point that stuck with me: edge compute is fixed, so agent harnesses built on elastic cloud assumptions don't transfer cleanly. Deciding what context to keep and how to compress it becomes a model problem. @liquidai
siliconangle.com/2026/10/08/liq…
siliconangle.com/2026/10/08/liq…
0 1 0 9 751 0
@liquidai reposted
Coming up in 5: Maxime Labonne from @liquidai (@maximelabonne).
Open d1, decision models that never write a word. Hopping on live.
youtube.com/watch?v=xFKnS3…
Open d1, decision models that never write a word. Hopping on live.
youtube.com/watch?v=xFKnS3…
0 4 0 5 2.2K 0
@liquidai reposted
Can't get enough of d1? Here are two more reasons to be excited:
- d1-3B: text + vision
- d1-omni-600M: text + (image or audio)
running these models locally means going beyond seeing the world to interacting with it in real time (shoutout to @NVIDIARobotics).
wild.
- d1-3B: text + vision
- d1-omni-600M: text + (image or audio)
running these models locally means going beyond seeing the world to interacting with it in real time (shoutout to @NVIDIARobotics).
wild.
0:42Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
2 2 0 27 1.4K 4
@liquidai reposted
Voice notes that file themselves 📥
d1-omni-600M sorts what you say into reminders, lists, messages, travel, questions or music in ~160 ms, entirely in your browser on WebGPU.
No transcript, nothing generated, just typed JSON. Your voice never leaves the tab.
Try it 👇
🎙️ Demo: hf.co/spaces/shreyas…
📦 ONNX (text + vision + audio): hf.co/onnx-community…
d1-omni-600M sorts what you say into reminders, lists, messages, travel, questions or music in ~160 ms, entirely in your browser on WebGPU.
No transcript, nothing generated, just typed JSON. Your voice never leaves the tab.
Try it 👇
🎙️ Demo: hf.co/spaces/shreyas…
📦 ONNX (text + vision + audio): hf.co/onnx-community…
0:40Liquid AI@liquidai · Oct 7d1-omni-600M is an experimental 600M-parameter model for text + image or text + audio.
It combines LFM2.5-Encoder-350M with vision and audio encoders, and leads our text benchmark comparison in toxicity detection and paraphrase identification.
Use it for voice-command routing, on-device moderation, and intent classification.
3/
It combines LFM2.5-Encoder-350M with vision and audio encoders, and leads our text benchmark comparison in toxicity detection and paraphrase identification.
Use it for voice-command routing, on-device moderation, and intent classification.
3/
Open quoted post →
2 10 2 96 5.6K 132
@liquidai reposted
Open weights decision models, will give it a spin
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
0 1 0 5 1.1K 1
@liquidai reposted
Openweight multi-modal decision models. From a premier American lab!
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
0 3 0 13 1.2K 0
@liquidai reposted
running d1-3B on a single RTX 3060. 12GB VRAM. (config below)
88.4% on BoolQ (full val, 3,270 questions), Liquid's card says 86.7. 44.5% on MMLU-Pro (all 12,032, random is 10%). ~43 ms p50 per decision, 21-23 decisions/s, 5 GB VRAM.
GPQA lands at 29.8%, close to random, which fits Liquid's own Decision Index showing knowledge as its weakest area.
added a decision harness to BenchKit for this. calibration next!
config (llama.cpp PR #30110, merged):
llama-server
-m d1-3B-Q8_0.gguf
--mmproj mmproj-d1-3B-F16.gguf
-ngl 99 -c 32768 -fa on
-np 1
-b 4096 -ub 4096
88.4% on BoolQ (full val, 3,270 questions), Liquid's card says 86.7. 44.5% on MMLU-Pro (all 12,032, random is 10%). ~43 ms p50 per decision, 21-23 decisions/s, 5 GB VRAM.
GPQA lands at 29.8%, close to random, which fits Liquid's own Decision Index showing knowledge as its weakest area.
added a decision harness to BenchKit for this. calibration next!
config (llama.cpp PR #30110, merged):
llama-server
-m d1-3B-Q8_0.gguf
--mmproj mmproj-d1-3B-F16.gguf
-ngl 99 -c 32768 -fa on
-np 1
-b 4096 -ub 4096
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
2 2 3 35 2.4K 15
@liquidai reposted
WILD. 2 days ago at SF tech week I said: at the frontier, there is no fundamental tech moat! no one has a superior tech or a new tech that is not achievable by others (this of course is a function of compute). the ultimate moat is fast & deep integration.
This means we are in the product-level competition. which is good. on the consumer side you see muse, instinct, grokbot, hark, and a 1000 more personal ai initiatives! everyone is doing the same thing! the best product wins.
on the enterprise side, the game is around neolabs/neoclouds selling inference and customization tokens in their market/sector, modality, and mode of choice.
to show you what I mean: decision models. Jev is a great idea, built over 2 years in stealth and launched closed. 2 weeks later we shipped d1, and today we open-sourced two d1 models, with recipes.
everything is gonna be open source, and everyone is gonna be on the same page.
This means we are in the product-level competition. which is good. on the consumer side you see muse, instinct, grokbot, hark, and a 1000 more personal ai initiatives! everyone is doing the same thing! the best product wins.
on the enterprise side, the game is around neolabs/neoclouds selling inference and customization tokens in their market/sector, modality, and mode of choice.
to show you what I mean: decision models. Jev is a great idea, built over 2 years in stealth and launched closed. 2 weeks later we shipped d1, and today we open-sourced two d1 models, with recipes.
everything is gonna be open source, and everyone is gonna be on the same page.
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
7 10 2 66 5.2K 23
@liquidai reposted
Super cool new decision models from @LiquidAI! These are like Jev, which is an emerging new class of AI models that make decisions in a single pass rather than generating tokens one by one.
But LiquidAI Open d1 are both open-weight and multimodal decision models running locally, making decisions in milliseconds!
The implications of decision models for robotics, autonomous laboratories, medical devices, and physical AI are staggering and very exciting! In particular, I very much like that they work with @nvidia Jetson at the edge, very relevant for robotics.
This will be the bridge of AI to physical world, where every task series are decision making for the next action.
But LiquidAI Open d1 are both open-weight and multimodal decision models running locally, making decisions in milliseconds!
The implications of decision models for robotics, autonomous laboratories, medical devices, and physical AI are staggering and very exciting! In particular, I very much like that they work with @nvidia Jetson at the edge, very relevant for robotics.
This will be the bridge of AI to physical world, where every task series are decision making for the next action.
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
6 7 0 81 10.3K 15
@liquidai reposted
👀 Jev can see now - Open source!
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
1 2 0 24 1.1K 4
@liquidai reposted
another decision model family, liquid's d1, and these are tiny
> d1-3b, text and vision, liquid says 8 ms on a 4090
> d1-omni-600m, text plus image or audio, no latency published yet
> one forward pass, odds over fixed answers, zero output tokens
> lfm open license, free under $10m revenue
clef flash 9b played breakout for me at 56 ms a move, 18.4 saves per ball, the 27b just sat under an idle paddle
a 3b at 8 ms should hold that paddle better, same harness, same game, i'll run both
> d1-3b, text and vision, liquid says 8 ms on a 4090
> d1-omni-600m, text plus image or audio, no latency published yet
> one forward pass, odds over fixed answers, zero output tokens
> lfm open license, free under $10m revenue
clef flash 9b played breakout for me at 56 ms a move, 18.4 saves per ball, the 27b just sat under an idle paddle
a 3b at 8 ms should hold that paddle better, same harness, same game, i'll run both
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
2 3 0 14 2.2K 4
@liquidai reposted
I have a Jetson AGX, so this collaboration between @liquidai and @NVIDIARobotics has my attention 👀
Open weight decision models running locally on Jetson, with a navigation demo in NVIDIA Isaac Sim.
Looking forward to getting Open d1 running on my own hardware!
Open weight decision models running locally on Jetson, with a navigation demo in NVIDIA Isaac Sim.
Looking forward to getting Open d1 running on my own hardware!
Liquid AI@liquidai · Oct 7We built 10 live-camera demos for d1-3B, from gesture-controlled games to content moderation, with one forward pass per frame.
In collaboration with @NVIDIARobotics, we also show d1-3B navigating an environment in Isaac Sim, with the model served on a Jetson in a hardware-in-the-loop setup.
5/
In collaboration with @NVIDIARobotics, we also show d1-3B navigating an environment in Isaac Sim, with the model served on a Jetson in a hardware-in-the-loop setup.
5/
0:42 1 2 0 16 1.1K 2
@liquidai reposted
Liquid AI just released two open-weight models for making decisions directly on your device.
d1-3B handles text and images. The experimental d1-omni-600M adds image or audio input at just 600M parameters.
A camera can flag a defect or a device can route a voice command locally, without a cloud round trip. Faster responses, less data leaving the device, and weights you can fine-tune yourself.
Pretty cool for small, practical AI systems :)
d1-3B handles text and images. The experimental d1-omni-600M adds image or audio input at just 600M parameters.
A camera can flag a defect or a device can route a voice command locally, without a cloud round trip. Faster responses, less data leaving the device, and weights you can fine-tune yourself.
Pretty cool for small, practical AI systems :)
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
24 15 1 242 19.8K 89
@liquidai reposted
Our first two open-weight decision models are out and they are multimodal! Decisions in just 8 ms on an NVIDIA GeForce RTX 4090.
Liquid AI@liquidai · Oct 7Today we release Open d1: two open-weight multimodal models in our d1 decision model family.
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
> d1-3B: text + vision
> d1-omni-600M: text + image or text + audio
> Real-time decision making anywhere, from data centers such as @nvidia DGX to RTX workstations to Jetson at the edge.
1/
Open quoted post →
1 3 0 25 1.7K 5