LFM2.5-2.6B runs at 30 tok/s on a phone

Liquid AI's LFM2.5-2.6B, a 2.69B-parameter model with 128K context and tool calling, runs at 30 tok/s on a phone via Q4_K_M GGUF. A custom engine achieves 17 tok/s on a OnePlus 13.