LiquidAI Drops Six LFM3 Edge Models: 1.7B Text With 256K Context Fits in 1GB at 4-bit
jacek2023 · reddit · 2026-10-28
LiquidAI released the LFM3 family — six models in one drop:
- LFM3-1.7B: a text model for local agents built on a new tri-hybrid architecture (13 gated short convolutions, 5 sliding-window attention, 6 global attention layers) with a 256K context window. At 4-bit it takes 1GB of memory, fitting any phone or laptop. Post-trained for harnesses like Hermes Agent, Pi, and OpenClaw; supports tool calling, structured outputs, and single-GPU fine-tuning. Not recommended for agentic coding or complex multi-step reasoning.
- LFM3-VL-2B: on-device vision-language model, the first LFM vision model to take video input, with step-by-step thinking and strengthened spatial/embodied reasoning as a robotics foundation.
- LFM3-Audio-2B: local audio model covering transcription, synthesis and translation across 18 languages, voice chat, audio reasoning, and speech-to-tool-calls; uses an in-house audio encoder and LiquidCodec tokenizer for better speech quality than LFM2.5-Audio.
- LFM3-Omni-2B: the family's first omni-modal model — one end-to-end model understanding text, image, video and audio, answering in text and speech, built for on-device use.
Related event: LiquidAI Releases LFM3 Family of On-Device Models(15 posts)→
More from Infra
- Hyperbolic raises $30M Series A1 to make fragmented GPU compute easier to use and trade — qinzytech · 2026-10-28
- Reflection AI runs 1.3B sandboxes in largest RL run, peaking at 883K vCPUs — zacharynado · 2026-10-28
- GPU capacity bust more likely than a bank run, argues AI compute cycle thread — bookwormengr · 2026-10-28
- Analyst: AI constraint map now tracks 50 components as everything hits bottlenecks — BenBajarin · 2026-10-28
- AXA Uses NVIDIA Earth-2 to Simulate 1,000+ Cyclone Seasons for Storm Risk Modeling — NVIDIAAI · 2026-10-28
- 2026 wafer fab equipment forecast raised to $153B as the cycle lengthens — BenBajarin · 2026-10-28