GLM 5.3 just outplayed GPT-5.6 Sol at its own game — for 20× less.
Соmmand Соde tested all three with the same /design prompt. GLM 5.3: $0.016 · playable GPT-5.6 Sol: $0.32 · nice UI, broken game Opus 5: $0.37 · best Subway Surfers-style clone GLM wasn’t the best overall. But
Karpathy fed opus 5 one paragraph from the lord of the rings
This isn’t AI generating a video It’s Opus 5 writing actual Three.js code - placing every polygon, calculating every character position, angle, and motion state, coordinate by coordinate, purely through logic What’s
“I’ll let you do an audit, see what we could improve, it’s Claude Opus 5 behind the code”
GPT 5.6 Sol :
I built a GTA game using Kimi K3 and I’m not impressed at all.
Compared to Gemini 3.7, GPT-5.6, Opus 5, Fable 5, and DeepSeek V4 Flash; Kimi K3 will need to do better
You have reached the end of the archive
All of Claude Opus 5