Qwen 3.8 27B running 2× faster on a local agentic harness is a pretty big deal.
Local agents live and die by iteration speed. If the model can reason, act, observe the result, and start the next step twice as fast, that adds up quickly over a long task.
Kimi K3 Qwen 3.8 Max
DeepSeek V4 Pro 0831 I am surprised that the performance of Qwen 3.8 Max is better than I expected. But it is the most expensive and takes a long time. DS V4 Pro failed.
Qwen 3.8 27b doing absurd things with ThreeJS.
If they weren’t labeled would you know which one was made locally on a 24gb card? Opus was still finishing when I cut it off at $5. New Gemini dirt cheap and not too bad, Grok solid. Not sure the local model lost here🤯
Qwen 3.8 just made everyone else look like they missed the prompt.
Bhavani_00007 tested Qwen 3.8, GLM 5.3, Grok 4.6, and Gemini 3.7 Flash on the exact same two-scene prompt. Qwen simply did what was asked. GLM came close. Grok messed up the coloring. Gemini fell behind. Qwen
You have reached the end of the archive
All of qwen38