I ran Qwen 3.8 27B, Nemotron 3.5 Lightning and Muse Glimmer on 16 hard problems.

One RTX 5090, Ollama, 64K context, Q4_K_M, and every model loaded alone on the GPU. Full write up with all the charts, the method, and the raw results:

KGP Talkie | Laxmi KantAug 141

Qwen 3.8 Max Full COURSE 1 HOUR (Build & Automate Anything)

Julian Goldie SEOAug 14

Same prompt with Qwen 3.8 27b.

Took 5x longer but its running on my macbook m5 max :)

SambitAug 141

GLM 5.3 had no business looking this good.

OmedTheVibeCoder tested GLM 5.3 Max against Fable 5, Qwen 3.8 Max, and Grok 4.6 on the same task. Fable 5 still takes it for me. But GLM 5.3 is suddenly much closer — and against Grok 4.6, I’d pick GLM pretty comfortably here. That’s

noclipepeAug 146