Measured 11 local LLM configurations. llama.cpp was too slow for Qwen3.8-Flash-Next, but with Strata and an NVMe SSD, it has become usable for daily tasks (though it's still slow...)“This model is ...