Qwen 3.8:27b - It's (maybe) not the new Messiah.....

Article automatically generated from technical news.

After a week of using Qwen 3.8:27b for semi productive work, here (at the risk of stating the bleedin' obvious) is my take: 1) It's (very) good for a small local model, but not without a caveat on VRAM required to make it useful IRL. 2) It uses (a lot of) thinking as a substitute for model size, which implies far more tokens are used in the context window (compared to other similar size models) 3) For agentic coding work in a limited context window, this means you will get less done be