Liquid AI releases LFM3: six 2B-class models on one backbone, led by a 1.7B text model with 256K context

Liquid AI has released LFM3, a family of six models in the 2B parameter range that share one backbone, LFM3-1.7B-Base. The family covers text, vision, audio, omni and decision models. According to Liquid AI, the 1.7B text model outperforms its predecessor LFM2.5-2.6B while using 35% fewer parameters. Liquid AI published the announcement on the Hugging Face blog on October 27, 2026.

The six models

Every LFM3 model branches from LFM3-1.7B-Base.

Model	Input	Output
LFM3-1.7B	Text	Text
LFM3-VL-2B	Text, images, video	Text
LFM3-Audio-2B	Text, audio	Text, speech
LFM3-Omni-2B	Text, images, video, audio	Text, speech
d1-2B	Text, images, video	Calibrated answers
d1-omni-2B	Text, images, video, audio	Calibrated answers

LFM3-Omni-2B is Liquid AI's first omni model, and LFM3-VL-2B is the first LFM vision model to accept video input. Liquid AI says every model can be fine-tuned on a single GPU.

Architecture and training

LFM3-1.7B-Base has 24 layers in a tri-hybrid mix: 13 gated short convolutions, 5 sliding-window attention layers with a 256-token window, and 6 global attention layers. The convolution kernel size is 7, up from 3 in LFM2. Other changes include head-wise gated attention, partial RoPE in the global attention layers, and sandwich normalization.

Liquid AI says its hardware-in-the-loop architecture search now covers 14 device and backend combinations, including phones, a Raspberry Pi 5, laptops, cloud CPUs, GPUs and the Qualcomm Hexagon NPU. According to the company, LFM3-1.7B matches LFM2-1.2B on short prompts per parameter across edge devices and is about 25% faster with 8K tokens in context.

The base model 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 more agentic traces and tool calls, which extends the context window to 256K tokens. The text model fits in about 1 GB at 4-bit precision, per Liquid AI.

Reported results

• LFM3-1.7B: Liquid AI reports 38.42 on WebArena-Verified, against 3.02 for LFM2.5-2.6B. It also says the model beats Qwen3.5-9B on long tasks in browsers and business apps. The model is post-trained for agentic harnesses such as Hermes Agent, Pi and OpenClaw.
• LFM3-VL-2B: It has 1.9B parameters against 3.1B for LFM2.5-VL-3B. Liquid AI reports the average across 62 vision benchmarks rising from 57.6 to 71.4, on par with Gemma 4 12B, which the company describes as more than six times larger.

New encoders and audio codec

The vision encoder combines convolution and attention. Liquid AI says it uses 3x less compute than the SigLIP2 encoder in earlier LFM VL models. The 164M-parameter encoder matches Qwen3.5-2B's vision encoder in accuracy with 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.

The audio encoder mixes sliding-window attention and short convolutions and is fully causal, so the model can prefill while the user speaks. Liquid AI cites a time to first token as low as 50 ms. LiquidCodec, the new audio tokenizer, turns generated tokens into speech at about 87x real time on a single CPU thread on an Apple M4 Max and supports streaming decoding.

Access

Each model runs in Hugging Face transformers, and LFM3-1.7B is published as LiquidAI/LFM3-1.7B (loaded with trust_remote_code=True). The model cards list other supported inference engines. Pricing not yet disclosed.

What this means

LFM3 puts text, vision, audio, omni and decision models on one shared 1.7B base, so teams working across modalities start from the same backbone rather than separate model lines. The reported WebArena-Verified score (38.42 against 3.02) and the 256K context are aimed at on-device agents, and the 4-bit footprint of about 1 GB makes that plausible on constrained hardware. The efficiency claims for the encoders and for the 25% speedup at 8K context come from Liquid AI's own measurements, so independent tests on target devices will show how they hold up outside Liquid AI's own evaluations.