LFM2-24B-A2B

A 24-billion parameter mixture-of-experts model from LiquidAI that activates only 2 billion parameters per token for efficient inference.

Strengths

Balances model capacity with computational efficiency through sparse activation, enabling faster inference than dense models of similar scale.

Best for

On-device deployment and latency-sensitive applications where reducing inference compute is important.

Limitations

May underperform dense models on reasoning tasks or domains where its training data is sparse, due to its smaller active parameter count.