Build specialized multimodal models with LFM3

Build specialized multimodal models with LFM3

Back to Articles

[0

Team

]

Article

Published October 27, 2026

[-1

Upvote

44

[

• +38

Leonie Monigatti

iamleonie

Follow

LiquidAI

Jimmy T.H. Smith

jimmysmith1919

Anthony Maltsev

amaltsev

Maxime Labonne

mlabonne

Brandon

Brandon3967

Samuel Stevens

samuelstevens

Ryan Shubert

shubeydoo

Tianshu Yu

tianshu-yu

Sina

s-jse

Tim Seyde

tseyde

Rouzbeh Amin Tafreshi

rouzbehat78

Alexander Amini

alexanderamini

Saniya Karwa

saniyakarwa

Marc Harkonen

haerski

Anna Piunova

orech

Calder Katyal

CalderKat

Ishwarya Ananthabhotla

ianant93

Farris Atif

farris-liquid

Tarek Dakhran

tdakhran

Paul Pak

paulpak58

xx

tugot17

Mihir Bafna

mihirbafna

Kyle Bruce Wheeler

m3m0ryh0l3

Aditya Tadimeti

adityatadimeti

Alfred Eng

alfredeng

Karthik Macherla

karthikrmacherla

Khoi Nguyen

KTLK

Aldo Gael Carranza

agcarranza

Aurelien Lac

Aurelien-Lac

Tyler Murray

undfined

Raahul Vignesh

Raahul26

Song Duong

sduong

Ramin Hasani

raminmh

Mathias Lechner

mlech26l

Akhilesh Deepak Gotmare

akhileshdeepakgotmare

Fernando Fernandes Neto

fernandofernandes

Jiahui Wang

WWWjiahui

Yuri Khrustalev

ykhrustalev

Alay Shah

alay2shah

Simon Lee

yungisimon

Mayank Agarwal

mayankagarwal

Zetian Li

zetianli

changyi yang

changyiyang

Sinoué GAD

GAD-cell

nathan ranchin

nathanrchn

LFM3 is our new generation of Liquid Foundation Models (LFMs), built to be the starting point for specialized intelligence. It brings text, vision, audio, omni, and decision models onto one shared backbone, all in the 2B parameter range. This means you can pick the model for your modality, fine-tune it on your own data on a single GPU, and deploy it where your application runs.

LFM3 brings four main changes compared to our previous LFM2 family:

• More capability per parameter: a new tri-hybrid architecture and training recipe allow LFM3-1.7B to outperform LFM2.5-2.6B while using 35% fewer parameters.

• Every modality on one backbone: six 2B-class models (text, vision, audio, omni, and decision) trained from LFM3-1.7B-Base, including our first omni model and video input for our vision model.

• Built for the edge: the text model fits in about 1 GB at 4-bit precision, new in-house vision and audio encoders speed up processing, and the text model’s 256K context window lets agents run on-device.

• Made to be specialized: every model fine-tunes on a single GPU, so you can adapt it to your data, terminology, and deployment environment.

How we built a more expressive backbone

LFM2 showed that a hybrid of short convolutions and attention can run fast on-device without sacrificing quality. For LFM3, we used our hardware-in-the-loop architecture search to redesign that architecture for agentic use cases, which need more capability per parameter and fast inference over long contexts. The search now covers 14 device and backend combinations, from phones and a Raspberry Pi 5 to laptops, cloud CPUs, GPUs, and the Qualcomm Hexagon NPU.

LFM3-1.7B-Base has 24 layers and four main changes:

• Tri-hybrid layer mix: 13 gated short convolutions, 5 sliding-window attention (SWA) layers with a 256-token window, and 6 global attention layers. The convolutions use a kernel size of 7, up from 3 in LFM2.

• Head-wise gated attention: a sigmoid gate scales each head's output, so a head can switch off when it has nothing to add. This can help suppress activation outliers that hurt quantization.

• Partial RoPE: global attention layers rotate only the highest-frequency half of each head, so the other half matches on content regardless of distance. SWA layers keep full RoPE.

• Sandwich norm: each block normalizes both its input and its output to keep activations well-scaled.

The payoff shows at long context. Per parameter, LFM3-1.7B matches LFM2-1.2B on short prompts across edge devices, and with 8K tokens in context it is about 25% faster. LFM3-1.7B-Base was pre-trained on 22T tokens with the same 128K vocabulary as LFM2.5-8B-A1B . The final 3.5T tokens are a mid-training stage on curated data with a larger share of agentic traces and tool calls, which also extends the context window to 256K tokens.

One backbone for six models

Every LFM3 model branches from LFM3-1.7B-Base. Instead of running one large general-purpose model, you pick the one that fits your inputs, customize it, and run it on-device.

Model

Input

Output

Built on

LFM3-1.7B

Text

LFM3-1.7B-Base

LFM3-VL-2B

Text, images, video

LFM3 backbone + vision encoder

LFM3-Audio-2B

Text, audio

Text, speech

LFM3 backbone + audio encoder, audio decoder, and LiquidCodec

LFM3-Omni-2B

Text, images, video, audio

LFM3 backbone + vision and audio encoders and audio decoder

d1-2B

Calibrated answers

d1-omni-2B

The vision and audio models also get two new in-house encoders:

• Vision encoder: it combines convolution and attention and uses 3x less compute than the SigLIP2 encoder in previous LFM VL models. Our 164M-parameter encoder matches Qwen3.5-2B's vision encoder in accuracy while requiring 2.2x fewer FLOPs (both measured with the same LFM2.5-230M language backbone). For video, it embeds each pair of sampled frames into one temporal slice, halving the temporal sequence length.

• Audio encoder and LiquidCodec: the audio encoder mixes sliding-window attention and short convolutions and is fully causal, so the model can prefill while you speak, with time to first token as low as 50 ms. LiquidCodec, our new audio tokenizer, turns generated tokens into speech at about 87x real time on a single CPU thread on Apple M4 Max and supports streaming decoding.

How to use the LFM3 models

Each model runs in transformers . The snippets below come from the model cards, which also list the other inference engines each model supports.

LFM3-1.7B

Reach for LFM3-1.7B when you need local agents, tool calling, structured outputs, or a base for domain-specific fine-tuning. It is post-trained to run in agentic harnesses such as Hermes Agent, Pi, and OpenClaw. It outperforms its predecessor, LFM2.5-2.6B, with 35% fewer parameters, and the biggest gains are on agentic tasks: on WebArena-Verified, it scores 38.42 against 3.02, and it even beats Qwen3.5-9B on long tasks in browsers and business apps. You can find more details in the LFM3-1.7B model card .

To run LFM3-1.7B, install the latest version of transformers :

pip install -U transformers

Then load and run the model:

from
transformers
import
pipeline

generator = pipeline(
"text-generation"
,
    model=
"LiquidAI/LFM3-1.7B"
,
    trust_remote_code=
True
,
)

messages = [{
"role"
:
"user"
,
"content"
:
"What is C. elegans?"
}]
output = generator(
    messages,
    do_sample=
True
,
    temperature=
0.1
,
    top_k=
50
,
    repetition_penalty=
1.1
,
    max_new_tokens=
2048
,
)
print
(output[
0
][
"generated_text"
][-
1
][
"content"
])

For local agent setup, see the Agent Harnesses guide .

LFM3-VL-2B

Reach for LFM3-VL-2B when you need on-device visual understanding. It is the first LFM vision model that takes video as input; it reasons step by step before it answers, and it finds objects and reasons about how they are placed, a strong foundation for robotics. Compared with LFM2.5-VL-3B, it has fewer parameters (1.9B against 3.1B) yet raises the average across 62 vision benchmarks from 57.6 to 71.4, on par with Gemma 4 12B, a model more than six times larger. You can find more details in the LFM3-VL-2B model card .

To run LFM3-VL-2B, install transformers>=5.21.0 with its vision dependencies:

pip install
"transformers>=5.21.0"
torch torchvision pillow

import
torch
from
transformers
import
AutoModelForImageTextToText, AutoProcessor

model_id =
"LiquidAI/LFM3-VL-2B"
model = AutoModelForImageTextToText.from_pretrained(model_id, dtype=torch.bfloat16, device_map=
"auto"
, trust_remote_code=
True
)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=
True
)

messages = [
    {
"role"
:
"user"
,
"content"
: [
            {
"type"
:
"image"
,
"url"
:
"https://placecats.com/300/200"
},
            {
"type"
:
"text"
,
"text"
:
"Describe this image."
},
        ],
    }
]

inputs = processor.apply_chat_template(
    messages,
    add_generation_prompt=
True
,
    tokenize=
True
,
    return_dict=
True
,
    return_tensors=
"pt"
,
).to(model.device)
with
torch.inference_mode():
    output_ids = model.generate(
        **inputs,
        do_sample=
True
,
        temperature=
0.1
,
        top_k=
50
,
        repetition_penalty=
1.1
,
        max_new_tokens=
16384
,
# generate() needs an explicit limit; reasoning can be long
)

generated_ids = output_ids[:, inputs[
"input_ids"
].shape[
1
] :]
print
(processor.batch_decode(generated_ids, skip_special_tokens=
True
)[
0
])

See the LFM3-VL-2B model card for tool calling, tool search, and on-device inference with llama.cpp.

LFM3-Audio-2B

Reach for LFM3-Audio-2B when you need voice chat, speech recognition, speech synthesis, or translation across 18 languages on a laptop or edge device. It takes text or audio as input and responds with text, speech, or both, and it can turn spoken requests into tool calls. Compared with LFM2.5-Audio-1.5B, it uses a new in-house audio encoder and speech codec, speaks with higher quality (UTMOSv2 3.83 against 3.22), and adds audio understanding, general-purpose audio reasoning, and tool calling. You can find more details in the LFM3-Audio-2B model card .

To run LFM3-Audio-2B, install transformers>=5.20.2 with its audio dependencies:

pip install
"transformers>=5.20.2"
torch accelerate librosa soundfile

Load the model and processor directly from the Hugging Face Hub:

import
torch
import
soundfile
as
sf
from
transformers
import
AutoModelForMultimodalLM, AutoProcessor

model_id =
"LiquidAI/LFM3-Audio-2B"
device =
"cuda"
if
torch.cuda.is_available()
else
"cpu"
dtype = torch.bfloat16
if
device ==
"cuda"
else
torch.float32

model = AutoModelForMultimodalLM.from_pretrained(
    model_id, dtype=dtype, device_map=device, trust_remote_code=
True
).
eval
()
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=
True
)
codec = processor.audio_tokenizer.to(device=device).
eval
()

Then ask a question aloud and get a reply in text and speech:

conversation = [
    {
"role"
:
"system"
,
"content"
:
"Respond with interleaved text and audio."
},
    {
"role"
:
"user"
,
"content"
: [{
"type"
:
"audio"
,
"audio"
:
"question.wav"
}]},
    {
"role"
:
"assistant"
,
"content"
:
"<|mixed_start|>"
},
]
inputs = processor.apply_chat_template(
    conversation,
    continue_final_message=
True
,
    tokenize=
True
,
    return_dict=
True
,
    return_tensors=
"pt"
,
).to(device)
with
torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=
1024
,
        audio_temperature=
1.0
,
        audio_top_k=
128
,
        audio_repetition_penalty=
1.1
,
        audio_repetition_window=
4096
,
        return_dict_in_generate=
True
,
    )

    text = processor.decode(
        outputs.sequences[
0
, inputs[
"input_ids"
].shape[
1
]:],
        skip_special_tokens=
True
,
    )
    waveform = codec.decode(
        outputs.audio_codes[:
1
, :, :outputs.audio_lengths[
0
]]
    ).audio_values[
0
]
# 24 kHz
print
(text)
sf.write(
"reply.wav"
, waveform.
float
().cpu().numpy(),
24_000
)

See the LFM3-Audio-2B model card for live speech recognition, speech synthesis, audio understanding, and tool calling.

LFM3-Omni-2B

Reach for LFM3-Omni-2B when you need one on-device model that sees, listens, and talks back, in place of a separate ASR model, vision-language model, LLM, and TTS model. It can answer spoken questions about an image, watch and listen to a video together, and turn what it sees and hears into tool calls. It is our first omni model, with text, vision, and audio input and both text and speech output, and it keeps the core capabilities of the standalone LFM3 models, including visual question answering, speech recognition, speech synthesis, and instruction following. Like LFM3-Audio-2B, it speaks through LiquidCodec, which turns generated tokens into speech faster than real time, so spoken replies stream as the model generates them. On audio-visual video understanding, it beats models up to 10.7B parameters, such as Qwen2.5-Omni-7B, on Daily-Omni, WorldSense, and AV-SpeakerBench. You can find more details in the LFM3-Omni-2B model card .

To run LFM3-Omni-2B, install transformers>=5.20.2 with its vision and audio dependencies:

pip install
"transformers>=5.20.2"
torch torchvision torchcodec accelerate pillow librosa soundfile

import
torch
import
soundfile
as
sf
from
transformers
import
AutoModelForMultimodalLM, AutoProcessor

model_id =
"LiquidAI/LFM3-Omni-2B"
device =
"cuda"
if
torch.cuda.is_available()
else
"cpu"
dtype = torch.bfloat16
if
device ==
"cuda"
else
torch.float32

model = AutoModelForMultimodalLM.from_pretrained(
    model_id, dtype=dtype, device_map=device, trust_remote_code=
True
).
eval
()
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=
True
)
codec = processor.audio_tokenizer.to(device=device).
eval
()

Then ask a question about a photo by voice:

conversation = [
    {
"role"
:
"system"
,
"content"
:
"Answer questions about the provided image. Base every answer only on what can be seen. Respond with text."
,
    },
    {
"role"
:
"user"
,
"content"
: [
            {
"type"
:
"image"
,
"image"
:
"photo.jpg"
},
            {
"type"
:
"audio"
,
"audio"
:
"question.wav"
},
        ],
    },
]
inputs = processor.apply_chat_template(
    conversation,
    add_generation_prompt=
True
,
    tokenize=
True
,
    return_dict=
True
,
    return_tensors=
"pt"
,
).to(device)
with
torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=
1024
, generation_mode=
"text"
)
print
(processor.decode(
    outputs.sequences[
0
, inputs[
"input_ids"
].shape[
1
]:outputs.sequence_lengths[
0
]],
    skip_special_tokens=
True
,
))

See the LFM3-Omni-2B model card for speech output, video with its soundtrack, and tool calling.

d1-2B and d1-omni-2B

Reach for the d1 models when you need fast, structured decisions instead of generated text. Give them a situation (a support ticket, an invoice, an app screen, a photo, a voice note) and a list of questions, and they answer all of them in a single forward pass, each with a calibrated probability. Beyond yes/no, choices, and scores, they can now point at objects, read values from documents, estimate numbers with a range, and pick every option that applies. You can find more details in the model cards for d1-2B and d1-omni-2B .

Between the two d1 models:

• d1-2B : choose it for text, pictures, and video. It replaces d1-3B with a third fewer parameters and scores higher on all four of our decision suites, for example, 75.5 against 64.0 on ImajevBench.

• d1-omni-2B : choose it when your inputs include audio. If you ask by voice instead of in writing, it keeps nearly all of its accuracy.

To run d1-2B, install transformers>=5.19 with its vision dependencies:

pip install
"transformers>=5.19"
torch torchvision pillow

Then load d1-2B and ask it several questions about one support ticket in a single pass:

import
torch
from
transformers
import
AutoModel

device =
"cuda"
if
torch.cuda.is_available()
else
"mps"
if
torch.backends.mps.is_available()
else
"cpu"
dtype = torch.float32
if
device ==
"cpu"
else
torch.bfloat16
model = AutoModel.from_pretrained(
"LiquidAI/d1-2B"
, trust_remote_code=
True
, dtype=dtype).to(device)
# Several named questions over one state, answered in one pass
questions = {
"refund"
: {
"type"
:
"noul"
,
"instructions"
:
"Is the customer asking for a refund?"
},
# yes/no: P(yes)
"team"
: {
"type"
:
"choice"
,
"instructions"
:
"Which team should handle this?"
,
"criteria"
: {
"billing"
:
"Charges, refunds, invoices"
,
"shipping"
:
"Delivery and parcels"
,
"technical"
:
"App or site faults"
}},
"topics"
: {
"type"
:
"choice"
,
"multiple"
:
True
,
"instructions"
:
"Which topics does the message raise?"
,
"criteria"
: {
"billing"
:
"Payments and charges"
,
"delivery"
:
"Shipping and parcels"
,
"login"
:
"Signing in to the account"
}},
"urgency"
: {
"type"
:
"score"
,
"instructions"
:
"How urgent is this ticket?"
,
"criteria"
: [
"low"
,
"medium"
,
"high"
]},
}
out = model.system_one(
"I was charged twice and my parcel never arrived. Please refund one charge."
, questions)
a = out[
"answers"
]
urgency = a[
"urgency"
][
"legend"
][
str
(
int
(a[
"urgency"
][
"score"
]))]
print
(
round
(a[
"refund"
][
"noul"
],
2
), a[
"team"
][
"choice"
], a[
"topics"
][
"choices"
], urgency)
# 1.0 billing ['billing', 'delivery'] medium

For brevity, we only include the example for d1-2B. See the d1-omni-2B model card for audio clips and spoken questions.

Fine-tune LFM3 for your task

LFM3 is built to be specialized. Frontier models are powerful generalists, but they are not necessarily experts in your use case. LFM3 packs extra capacity into a small, fast model, so a short fine-tune can turn it into a reliable expert on your task, with your data, terminology, and tools. Because it is small, you can deploy the fine-tuned model where your application runs, from devices to private infrastructure to the cloud.

Every model can be fine-tuned on a single GPU, and these notebooks walk you through a complete task for the text, vision, audio, and omni models, from data preparation to evaluation.

Method

Task

Notebook

SFT

Extract structured JSON from messy documents

Open in Colab

DPO

Answer from retrieved documents, or decline

GRPO

Call the right tools in a 108-tool phone app

CPT

Learn a new language (Turkish) from raw text

Read receipt photos into JSON

Transcribe images of equations to LaTeX

Map spoken smart-home commands to function calls

Teach the model a new voice

Answer spoken questions about medical images

Example: teaching LFM3 to act

We adapt LFM3-VL-2B into three complementary robotics models for orchestration, navigation, and manipulation . The orchestrator operates a robot arm through tool calls: it grounds objects by pointing, measures them with depth tools, invokes motion planners, and checks outcomes in subsequent camera images. The navigator follows route descriptions using recent RGB observations, predicting spatial targets and a waypoint trajectory in just five tokens. The manipulation VLA maps images, language, and robot state to eight-step chunks of continuous actions in a single forward pass. Jointly trained across 15 embodiments, the VLA achieves 96.2% average success across the four LIBERO suites and 45.7% on RoboCasa-GR1 tabletop tasks, generating action chunks in under 50 ms on a Jetson AGX Orin with TensorRT FP16. These results highlight LFM3-VL-2B as a compact foundation for robotics, spanning high-level planning to low-level control with real-time action generation on embedded hardware.

Watch LFM3-VL-2B handle perception, navigation, and manipulation in this demo video .

Get Started with LFM3

All LFM3 models are open-weight and available on Hugging Face today:

• Download: the LFM3 collection on Hugging Face, including LFM3-1.7B-Base for your own post-training.

• Fine-tune: adapt LFM3 to your task with our fine-tuning notebooks .

• Use in your harness: follow our guide on how to run a local agent, like OpenClaw, Hermes Agent, and Pi.

We can't wait to see what you build.

Citation

If you use this work, please cite the release blog:

@article{liquidAI2026LFM3,
  author = {Liquid AI},
  title = {Introducing LFM3: Every Modality for Specialized Intelligence},
  journal = {Liquid AI Blog},
  year = {2026},
  note = {www.liquid.ai/blog/lfm3},
}

Models mentioned in this article 8

Collections mentioned in this article 1

More from this author

Run fast, private agents on-device with LFM2.5-24B-A2B

70

October 13, 2026

Multimodal open d1 decision models for the edge

43

October 7, 2026

Community

Edit

Preview

Upload images, audio, and videos by dragging in the text input, pasting, or

clicking here

.

Tap or paste here to upload images

Comment

· Sign up or log in to comment

• +32