Liquid AI · LFM2.5
LFM2.5-2.6B
A 2.6B text model from Liquid AI for on-device agents, with vendor measurements on phone, Apple and AMD hardware.
- Parameters
- 2.6B
- Context
- 128,000 tokens
- Input
- Text
- Output
- Text
- Access
- Open weights
- Licence
- Model-specific terms
- Released
- Device evidence
- Measured on device
- Last verified
- Formats
- BF16, GGUF, ONNX
- Runtimes
- llama.cpp, ONNX Runtime
- Platforms
- Android, Linux, macOS, Windows
Liquid AI reports 30 tokens per second and 2,448 MB of memory for a Q4_K_M build on a Snapdragon phone. These are vendor measurements rather than independent results.