Liquid AI releases LFM3, small open models tuned for Hermes Agent and OpenClaw
Announced · Published · By Contractor AI Shop
Liquid AI released LFM3, six ~2B-class open-weight models on a shared 1.7B base. The text model fits in about 1 GB at 4-bit, has a 256K context, and is post-trained for agent harnesses including Hermes Agent, Pi and OpenClaw.
Key facts
- Liquid AI says LFM3 includes six models (text, vision, audio, omni, and two decision models) sharing a 1.7B foundation, with the text model supporting a 256K context locally. Liquid AI
- Liquid says the text model is post-trained to run in agent harnesses including Hermes Agent, Pi and OpenClaw. Liquid AI
- VentureBeat (Carl Franzen) reported Liquid's own benchmarks and noted larger models still win on agentic coding and knowledge; Liquid docs say it is not recommended for agentic coding. VentureBeat (Carl Franzen)
- MarkTechPost (Asif Razzaq) covered the open-weight multimodal LFM3 release. MarkTechPost (Asif Razzaq)
Our take: what it means for contractors
A cheap always-on background agent candidate for local setups, not a replacement for a big daily model. Liquid's own table shows larger models still win on coding and knowledge.
What we don't know yet
- Independent agent tests
- Full local memory budget at 256K context
- Commercial license terms above $10M revenue (LFM Open License v1.0)
Sources
- Liquid AI · LFM3 · 2026-10-27 · Primary
- VentureBeat (Carl Franzen) · Liquid AI’s new LFM3 models put browser agents, voice commands, and image recognition onto enterprise-controlled devices · 2026-10-27 · Secondary
- MarkTechPost (Asif Razzaq) · Liquid AI LFM3: Open-Weight 2B Multimodal Models · 2026-10-27 · Secondary
Related
- OpenClaw
- Hermes Agent (Nous Research)
- oMLX
- Local AI news
- Local AI tools
- AI Agents news
- AI Agents tools
- Latest weekly roundup