From the source
Liquid AI released LFM2.5-24B-A2B, an open-weight 24B-parameter reasoning model with 2B active parameters, on October 13, 2026.
The model is built for on-device agents and features a 131,072-token context window, a 128,000-token vocabulary, and support for 16 languages.
It leads or matches comparison models on five of eight tool-use and agentic benchmarks and achieves decode speeds of 184 tokens per second on an Apple M5 Max.