Liquid AI releases open-weight LFM3 model family with 1.7B text model
Article timeline
The articles in this story. Times are ET.
ModelScopeOfficialLFM3-1.7B-GGUF on ModelScope: 256K context, 1 GB at 4-bitAIModelScope now hosts LFM3-1.7B-GGUF, a 1.73B-parameter model with a 256K context window that uses about 1 GB at 4-bit precision, according to Atomic Chat, which made the GGUF conversion of Liquid AI's model. Liquid AI's reported 32K-context evaluation shows at least 2.5x faster prefilling and 1.8x faster decoding than every compared model across three tested devices. The GGUF files range from 1.85 GB (Q8_0) to 0.92 GB (IQ3_S), and the publisher reports Q8_0 keeps 96.3% top-1 agreement with the reference model.
LocalLLMXLocal LLM adds LFM3 1.7B to its iOS app for on-device useAILocal LLM says LFM3 1.7B is now available in its iOS app and that it can build a small interactive webpage on an iPhone. The author also reports that generation sped up. The quoted Liquid AI announcement describes LFM3 as a model family spanning text, vision, audio, and omni modalities, with a 1.7B text model that has a 256k context window.
- MarkTechPostNewsLiquid AI releases LFM3, six open-weight 2B-class multimodal models
AILiquid AI has released LFM3, a family of six open-weight 2B-class models built on one shared backbone, covering text, vision, audio, omni and decision tasks. LFM3-1.7B runs in about 1 GB of memory at 4-bit and posts 49.37 across 29 benchmarks in Liquid's own evaluation, ahead of Qwen3.5-9B at 47.24. Knowledge and agentic coding remain weak spots, and commercial use is free only below $10M in annual revenue.
ollamaOfficialOllama adds Liquid AI's LFM3 small models for local useAIOllama says Liquid AI's LFM3 is now available to run locally, with text, vision, audio, omni, and decision models. The text model fits in about 1 GB at 4-bit and has a 256K context window. The models are built to run anywhere and fine-tune on a single GPU.
Rohan PaulXLiquid AI launches LFM3 open-weight models, with a 1.7B text model beating its 2.6B predecessorAILiquid AI has launched LFM3, an open-weight model family, and the author says its 1.7B text model beats the 2.6B LFM2.5 predecessor. The gain is attributed to a tri-hybrid design of 24 layers, with 13 short convolutions, 5 sliding-window attention layers and 6 global attention layers. On AutomationBench, Liquid reports 41.29 for the 1.7B model against 15.10 for Qwen3.5-9B.
elvisXPickLiquid AI releases LFM3, a family of six open-weight models for on-device tasksAILiquid AI has released LFM3, a family of about 2B open-weight models covering text, vision, audio, omni, and decision tasks on one shared backbone. The author says LFM3-1.7B supports a 256K context and scores 38.4 on WebArena-Verified, versus 28.9 for Qwen3.5-9B in Liquid's own evaluation. The models can be fine-tuned on a single GPU.
SantiagoXLiquid AI releases the LFM3 family of small open-weight modelsAILiquid AI introduces LFM3, a family of six open-weight models for text, vision, audio, omni, and decision tasks. Most variants are about 2B parameters, and the source says they can be fine-tuned on one GPU and run on-device. Liquid AI also claims LFM3-VL-2B performs competitively with Gemma 4 12B, a model six times larger, and that LFM3-1.7B ranks first on the Pareto frontier of device intelligence.
LM StudioOfficialLM Studio adds Liquid AI's LFM3 models for local useAILM Studio says LFM3 is now available in its app, and the model fits in about 1 GB of memory. The model page is at The quoted Liquid AI post describes the LFM3 family as six instances across text, audio, vision, and omni modalities, including the 1.7B text model with a 256k context window.
- Hugging Face BlogOfficialPickLiquid AI releases LFM3, a family of 2B-class multimodal models on one backbone
AILiquid AI releases LFM3, a family of six 2B-class models for text, vision, audio, omni, and decision tasks, all built from LFM3-1.7B-Base. The company says LFM3-1.7B outperforms LFM2.5-2.6B with 35% fewer parameters, and the text model fits in about 1 GB at 4-bit precision with a 256K context window. All models are open-weight on Hugging Face and can be fine-tuned on a single GPU.
SGLangOfficialSGLang adds day-0 support for Liquid AI's LFM3 small modelsAISGLang says it supports Liquid AI's LFM3 family on day zero, with a specialized small-model suite that can run on the edge. The post says LFM3-1.7B beats LFM2.5-2.6B with 35% fewer parameters, and that it fits in about 1 GB at 4-bit with a 256K context. A sample sglang serve command for LiquidAI/LFM3-1.7B is shown, and the models can be fine-tuned on a single GPU.
Liquid AIOfficialLiquid AI releases LFM3, a family of six multimodal foundation modelsAILiquid AI introduces LFM3, a new generation of Liquid Foundation Models with a tri-hybrid architecture and new multimodal encoders. The family has six instances across text, audio, vision, and omni modalities, including LFM3-1.7B with a 256k context window and LFM3-VL-2B with video input. The post says the models can be fine-tuned on one GPU and run on-device.
