There is minimal downside to switching to open models

… if you have the ram (with fast enough bandwidth)

MoE models are pretty magic on my laptops 32gb ram. 24 tok/sec on DDR5-5600 using Gemma 4 26B-A4B is so much faster than a dense model