LFM3 (1.7B, Q4)
Liquid AI's LFM3 at 1.7B parameters. A tri-hybrid convolution-attention design built for local agents — about a gigabyte of memory, quick even on modest hardware, with integrated reasoning, native tool calling, and a 256K context window.
| At a glance | |
|---|---|
| Made by | Liquid AI |
| Download size | 1.0 GB |
| Parameters | 1.7B |
| Quantization | Q4_0 |
| Context window | 262K tokens |
| Uses tools | Yes |
| Best for | Everyday work |
| Runs on | llama.cpp |
| License | LFM Open 1.0 |
| Added to the catalog | 2026-10-31 |
Gezel downloads this model the first time a gezel needs it, checks the file against its published hash, and applies tuned settings for it.
Measured results
Not measured yet. No scorecard round has run this model or another build of it, which is not a verdict either way. How we test models explains what a round measures, and you can run the same tests on your own machine:
gezel bench --model lfm3-1.7b-q4