How to run Qwen 3.8 27b (Opus 4.6) At Home 🏆
Pushing 30+ tk/s on Mac M2 Max // 97 tk/s on 4090 TL;DR: Using Codex + GPT‑5.6 SOL at xHigh reasoning, I stress-tested the new Qwen3.8‑27B across my local hardware and nearly doubled its speed through native speculative decoding.
Qwen3.8-27B fits in 24GB.
The agent that scored 73.0 on Terminal-Bench does not. Qwen dropped 3.8-27B yesterday under Apache 2.0 and the number everyone is repeating is the community Q4 GGUF at roughly 17GB, which makes a 24GB card sound like enough. I have spent four days
Qwen 3.8 27B is an absolute beast considering its intelligence density
There's simply no question about that. For those of you struggling with thinking verbosity, I have put together a video comparison below showcasing an HTML canvas test across a sweep of thinking budgets! I
Qwen 3.8 27B just ran 2× faster on local agentic harnessing
You have reached the end of the archive
All of qwen38