Running Qwen 3.8 27B Q4 on a single RTX 4060 (8GB VRAM) gaming laptop shouldn’t be physically possible.
You can now run this 27B dense beast locally with a massive 64,000 token context window (minimum requirement by hermes agent) on budget consumer hardware, without crashing
Alibaba’s Qwen team just released a genuinely usable open model, not another oversized preview nobody can run.
Qwen 3.8 27B, Apache 2.0 license — fully open, commercial use included. 27B parameters with vision built in, 262K native context extensible to 1M. The FP8 variant
Qwen 3.8 27b one-shot meteor game with 4bit unsloth quant so it works comfortably in 24GB RTX 4090.
I was trying to answer the question: "is qwen 3.8 good enough for local coding?" I didn't focus on tok/s because you can hack that with small contexts. My demo was against
Qwen 3.8-27B: local AI just closed a 2-year gap in 6 months.
Claude Opus 4.6 was the best model in the world 6 months ago. Now a free model matching it runs on YOUR machine. → Open weights, download it tonight → Runs on a Mac Mini or an RTX card (no crazy GPU needed) →
You have reached the end of the archive
All of qwen38