Running Qwen 3.8 27B on 2×5090s with a few basic optimizations
~230 tokens/s (video is real-time) two years ago building a pubsub would have taken me an entire day now it takes 8 seconds
WARNING: Next week could be crazy for AI.
A list that is circulating points to 11 possible new models and updates: 🤖 Claude Fable 5.1 🚀 GPT-6 Astra ⚡ Grok 4.7 🇨🇳 GLM-5.3 Flash 🔓 Qwen 3.9 Open Weights 🎬 Composer 3 🧠Ox Alpha Flash ✨ Gemini 3.8
Yesterday I told you two examples of agentic use where Qwen 3.8 27B executed locally had done clearly better than Gemini 3.7 flash.
In this video is the second of them. I gave both models access to a folder with 42 disorganized documents from the activity of a
Build a Construction Site simulator scene using 0x Alpha and compared it with Grok 4.6 and Qwen 3.8
You have reached the end of the archive
All of qwen38