If you believe a 128gb machine that is essentially DGX Spark in a laptop chassis can run models comparable to SOTA you either never ran open models on hard tasks, or you aren't scratching the surface of SOTA closed LLM capability in how you're using them.