LFM2.5-24B-A2B

Summary from the TLDR AI newsletter

LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters, is designed for fast, private on-device agents. It outperforms or matches larger models like Qwen3.6-35B-A3B on five out of eight benchmarks and runs up to 1.7x faster due to its efficient architecture. Available on Hugging Face, it supports extensive inference frameworks, making it suitable for tasks needing swift local inference and tool usage.