LFM3-Omni-2B-MLX-4bit Multimodal Model: 4GB+ VRAM

LFM3-Omni-2B-MLX-4bit Multimodal Model: 4GB+ VRAM

At a Glance

Item Value
Repository LiquidAI/LFM3-Omni-2B-MLX-4bit
Family guide LFM3-Omni guide (1 articles)
Publisher guide Liquid AI: models and licenses
Published 2026-10-27
License lfm1.0
Formats MLX / safetensors
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

LiquidAI has released “LiquidAI/LFM3-Omni-2B-MLX-4bit", a 4-bit quantized model optimized for Apple Silicon. Based on the omni-modal model “LiquidAI/LFM3-Omni-2B", which supports mutual understanding and generation of audio, vision, and text, it is designed for efficient operation on Apple Silicon-powered devices. The model weight file size is 1.96GB, with an additional 0.15GB required for audio codec data.

Specifications

  • Parameters: 2.17B total parameters (base model specifications)
  • Architecture: LFM3 backbone, LFM3 vision encoder, LFM3 causal audio encoder, LiquidCodec audio tokenizer
  • Context length: 32K tokens (combined input and output)

Performance

Based on the measurement scores published by the creators of the base model “LiquidAI/LFM3-Omni-2B" in its model card, we introduce a performance comparison with other models. Note that all performance metrics introduced in this article are for the base model, and the 4-bit quantized version may experience a decrease in accuracy due to quantization.

First, the evaluation results for automatic speech recognition (ASR) and text-to-speech (filtered to include selected comparison models) are as follows:

→ Scroll horizontally to see all columns

Benchmark LFM3-Omni 2B Qwen2.5-Omni 3B Gemma 4 E2B Gemma 4 E4B Phi-4 multimodal
Total parameters 2.2B 6B 5B 8B 5.6B
English ASR WER ↓ 6.36 6.30 8.26 7.82 5.14
Multilingual ASR WER / CER ↓ 6.72 7.41 8.08 6.54 — †
English TTS UTMOSv2 ↑ 3.80 3.42 — — —
VoiceBench ↑ 64.73 66.37 63.28 72.90 64.97
Tool calling internal eval ↑ 52.20 44.30 55.60 72.40 —
Audio understanding MMAR ↑ 51.40 54.41 44.49 49.10 44.69

In the field of speech recognition, it recorded an English ASR word error rate (WER) of 6.36, showing a level comparable to Qwen2.5-Omni 3B. It also scored higher than Gemma 4 E2B in multilingual ASR error rates. Furthermore, it scored 3.80 in English TTS UTMOSv2, which measures speech synthesis quality, showing superiority among the comparison targets.

Next, the benchmark results for image and vision-related tasks are as follows:

→ Scroll horizontally to see all columns

Benchmark LFM3-Omni 2B Qwen2.5-Omni 3B Gemma 4 E2B Gemma 4 E4B Phi-4 multimodal Cosmos3 Edge
Total parameters 2.2B 6B 5B 8B 5.6B 4B
General
MMStar 55.67 55.67 46.47 52.00 59.33 56.07
MMMU (val) 42.44 48.00 37.81 43.14 54.76 37.14
MMMU-Pro 27.28 24.80 24.34 28.67 27.63 13.76
MMBench 73.68 79.30 67.61 73.54 82.73 76.63
SEEDBench (image) 75.07 75.27 72.53 74.78 72.24 74.73
RealWorldQA 66.54 63.14 48.63 53.07 64.84 72.81
MathVista (mini) 62.50 58.80 51.00 54.30 65.40 53.90
POPE 89.34 84.62 78.78 81.24 83.34 83.52
Documents and OCR
DocVQA 92.47 92.71 73.49 77.91 92.60 84.43
InfoVQA 64.08 74.55 38.72 48.30 65.98 60.94
OCRBench 806 863 731 761 795 751

In the field of vision and image understanding, it records scores comparable to models with larger parameter counts, marking particularly high figures in items such as POPE and DocVQA. On the other hand, in some advanced college-level image understanding tasks like MMMU (val), it falls short of larger-scale models.

Strengths and Use Cases

The base model, LiquidAI/LFM3-Omni-2B, is an omni-modal model capable of simultaneously processing text, images, audio, and videos with audio tracks. It supports a diverse range of tasks including automatic speech recognition (ASR), text-to-speech (TTS), voice chat via interleaved audio and text output, image and video question answering, and translation. It is also equipped with audio-based tool calling capabilities, designed with smart phone voice assistants and agent use cases in mind. Supported text languages include English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, and Polish.

Hardware Requirements

Estimated requirements (calculated by Local Model Watch) — 2.2B parameters (taken from the base model LiquidAI/LFM3-Omni-2B)

Your VRAM Quantization File size Est. memory needed
4GB (laptop iGPU / phone class) MLX 4bit 1.8GB 2.2GB

Inference engine support (architecture name matched against each project’s own model registry in its source code, checked 2026-11-17): llama.cpp: registered, vLLM: not registered, MLX (mlx-lm): not registered. “Not registered" means the name is absent from that registry today, not that the model cannot run.

Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

Runs in Ollama, LM Studio and llama.cpp via a converted build.

The publisher ships MLX / safetensors, but ggml-org/LFM3-Omni-2B-GGUF provides a GGUF build you can use.

License — lfm1.0 (Commercial use allowed with conditions): LiquidAI’s LFM Open License v1.0. Commercial use is allowed, except by legal entities with annual revenue of $10 million or more (Section 5). Redistribution requires the license text and notice of changes.

Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists. License summaries are not legal advice — check the publisher’s original terms before relying on them.

How to Get It

This model is available as an MLX format file from the Hugging Face repository. To use it, download it from the Hugging Face page.

  • Distribution format: MLX, safetensors
  • Example download command:
python -m mlx_vlm.generate --model LiquidAI/LFM3-Omni-2B-MLX-4bit --temperature 0 --system "Perform ASR in English." --prompt "" --audio speech.wav
  • Supported engines: MLX-VLM

Note that according to the creators’ model card, this model is a minimal 4-bit quantization variant and therefore suffers from some loss in accuracy. For higher accuracy, the 8-bit version (LiquidAI/LFM3-Omni-2B-MLX-8bit) is recommended.

Quantized and Converted Variants

→ Scroll horizontally to see all columns

Added Publisher Format Repository Smallest VRAM tier (build, est. memory)
2026-10-28 ggml-org GGUF ggml-org/LFM3-Omni-2B-GGUF BF16 3.9GB (fits in 4GB VRAM)

File sizes of each build:

  • Available builds in ggml-org/LFM3-Omni-2B-GGUF: Q4_K_M 1.0GB / Q8_0 1.7GB / BF16 3.2GB

This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files. What the quantization names mean: glossary.

Related Articles

What to Read Next

Sources