I tried to look at the paradoxes of these models
Gemini 3.7 Flash, Qwen 3.8, Grok 4.6, and GLM 5.3 are all on the scene now. 🤔
We just upgraded from qwen 3 8b → qwen 3.8 27b by OrcaRouter, finally bringing you private & uncensored ai, with 3.4x the parameters and 128k context.
A 27B model on local hardware just beat claude opus 4.6
Qwen 3.8 27b, 4bit quant on a single gpu, out-built opus 4.6 on a minecraft clone in claude code. 27 billion parameters against a trillion. and it won the crazy part is the reasoning levels. low effort made the best
What's CRAZY about llms?
Instead of searching for simple no-harness chat with html preview... I just asked qwen to deliver. 1 coffee later I had my simple app ready - custom made to my tasks. Love it 🩷 Also: 3.8 has some nice webdev skills :)
You have reached the end of the archive
All of qwen38