Liquid AI releases LFM3: 2B open-weight multimodal family with 256K context
omarsar0 · x · 2026-10-28
Liquid AI launched LFM3, a new generation of Liquid Foundation Models built on a tri-hybrid architecture with new multimodal encoders, spanning six 2B open-weight instances across text, vision, audio, omni, and decision tasks on one shared backbone.
- LFM3-1.7B: text model with 256K context, trained to run in agent harnesses like Hermes Agent and Pi; scores 38.4 on WebArena-Verified vs. 28.9 for Qwen3.5-9B in Liquid's own eval
- LFM3-VL-2B: vision-language model with video input for spatial understanding
- LFM3-Audio-2B: tool calls, voice chat, reasoning in 18 languages
- LFM3-Omni-2B: combines vision, audio, and language
- d1-2B / d1-Omni-2B: decision models emitting typed outputs with calibrated probabilities in a single pass
All models can be fine-tuned on a single GPU, making them strong candidates for always-on local agents.
More from coding & agent
- Open-source agent memory layer hits 86.67% on LongMemEval without embeddings — GabrielMartinMoran · 2026-10-28
- Claude Opus 5.5 scores 18.5% on ProgramBench, up from 0% at May launch — winer666 · 2026-10-28
- Sundar Pichai: I run agents that summarize Google's training updates and compute telemetry for me — rohanpaul_ai · 2026-10-28
- Amazon rewrites SDE role guidelines for the 'Renaissance Developer' era of AI coding — prdeepakbabu · 2026-10-28
- Kinocut plugin lets your Hermes agent edit video locally with FFmpeg — no account or API key — alexcovo_eth · 2026-10-28
- Open-source agent memory layer Mind hits 86.67% on LongMemEval without embeddings — GabrielMartinMoran · 2026-10-28