Is It Possible To Run Qwen 3.8 Max Locally?

Qwen 3.8 Max costs $2/M tokens via API, but running this 2.4T-parameter mixture-of-experts model locally needs terabytes of VRAM Qwen3.8-Max is Alibaba's new open-weight frontier model: 2.4 trillion total parameters in a mixture of experts architecture, with only about 95 billion active per token. That gap between active and total parameters is the whole story, it makes the API cheap to run ($2 per million input tokens, $6 output) while making local hosting a multi-terabyte hardware problem, since every expert has to sit resident in VRAM even though most never fire on a given token. This video breaks down exactly why that architecture creates two separate bills, a cheap speed bill and a brutal memory bill, and does the actual math on what it would take to hold this thing in memory yourself. It walks through Nvidia H100 pricing and rental costs, why renting a GPU cluster just re-creates the same price you were trying to avoid, and why quantization can't save you when a 4x cut to a multi-terabyte model still leaves you with a multi-terabyte model. It also covers where Qwen3.8-Max actually lands on independent benchmarks (unimpressive overall, #1 in reasoning), and the much smaller Qwen3-8B sibling you can actually run today on a single consumer GPU with a basic Ollama command. For anyone tracking open weight models, mixture of experts architecture, or the real economics of self-hosting a frontier-class transformer architecture model, this is a grounded look at where the download-it-yourself dream currently falls apart, and where it doesn't. Built for builders and the AI-curious who want to know what it actually costs, in dollars and silicon, to own a model instead of renting one. Chapters: 0:00 The $2 Model With A Hidden Bill 0:22 What Owning It Actually Gives You 1:40 Why It's Cheap To Run 3:21 Why It's Brutal To Store 4:35 The House-Sized Hardware Problem 6:30 Can You Just Shrink It 8:16 The Door That's Actually Open 9:52 Is The Flagship Even Worth It 11:20 Can You Actually Get It Yet Tools & resources mentioned: - Qwen3.8-Max: https://www.yottalabs.ai/post/qwen-3-8-max-release-date-specs-how-to-access-2026 - Qwen3-8B - Ollama: https://ollama.com - Hugging Face: https://huggingface.co - KTransformers - DeepSeek-V3/R1 - Nvidia H100 About The Stack The Stack helps you build with AI. Each video takes one tool, model, or workflow and shows how it works in a few focused minutes, with the real benchmarks and real costs. We go deep on Claude Code and Cursor for AI coding, AI agents and MCP servers, the open-source AI tools and GitHub repos most people miss, RAG and vector search, fine-tuning, and running local LLMs on your own machine with Ollama and LM Studio. We compare models like ChatGPT and Claude, test AI automation with Zapier, Make, and n8n, and flag the tools that actually ship. Subscribe for new breakdowns: https://www.youtube.com/@the-stack-ai?sub_confirmation=1 #qwen #aiagents #openweightmodels #mixtureofexperts #localllm