M5 Max 128GB, 17 models, 23 prompts: Qwen 3.5 122B is still a local king

A tech professional has demonstrated a powerful real-world application for local large language models (LLMs) by using the Qwen 3.5 122B model on a new M5 Max MacBook Pro with 128GB of unified RAM. The user highlights the stark contrast in open model output, noting that while US-based releases like Meta's Llama, IBM's Granite 3.3, and Google's Gemma 4 have been rare, Chinese models from Qwen, DeepSeek, and others have been consistently solid. The key innovation enabling this test was bringing compute power home with the high-spec MacBook, moving away from rented GPUs on Google Cloud or AWS for the primary benefit of data privacy.

The specific use case involves solving a chaotic organizational problem: managing disparate data from US public school systems for his children. The workflow involves scraping data from various school-affiliated websites—some with APIs, others requiring tools like Playwright to navigate—where information is buried in slide decks, obscure folders, and disconnected systems. The Qwen 3.5 122B model is tasked with scouting this ambiguity to extract clear signals about due dates, grades, and assignments. This proves an ideal LLM application, transforming unorganized text into actionable insights while ensuring sensitive family data never leaves the local device, unlike cloud-based alternatives from OpenAI, Anthropic, or Google.