SYLIN

LFM

by Liquid AI · Model · Language & code models

LFM3-1.7B is a tri-hybrid model for local agents and specialized applications.

AccessOpen weightsPrice$0.08 in · $0.2 out per M tokensContext262K tokensNeeds~3.5 GB; smallest build ~1 GBDownload3.5 GBLicenseApache-2.0 (mostly)

Versions

VersionReleasedRank
LFM 2.5LATESTOct 18
LFM 3

Evidence

  1. Nov 1
    Liquid AI's ~2B family: text, vision, audio, omni and decision models, in GGUF, FP8, FP4 and MLX; text base ~3.5 GB
  2. Oct 18
    LFM2.5-24B-A2B: 24B MoE with 2B active, 48 GB, mostly Apache 2.0, built for high-end consumer devices
  3. Oct 18
    Claimed on average stronger than similar-sized Qwen 3.6 and Gemma 4 on agentic coding, tool use and knowledge, and faster

Cost to run

Alternatives

catmindScores 24.3% on math versus 75.6% for LFM2.5 1.2B Thinking, a 51-point drop from replacing the trace content

Builds by others

9
Show all 9