Key Info
Liquid AI's LFM2.5-24B-A2B model scores 96.75 on IFEval, using only 2.45B active parameters out of 24B total, placing it first on the Papers with Code instruction-following leaderboard.
Highlights
- IFEval score of 96.75, the top result on Papers with Code's instruction-following leaderboard
- Only 2.45B active parameters out of 24B total, indicating a sparse/MoE-style architecture
- Outperforms Rufus-Air (95.4) and Qwen3.5-27B (95)
- Highlights efficiency gains in instruction following relative to larger dense models