
I've tested 3 open llm models on Alibaba cloud to build a simulation game. With opencode.
Results:
```
MiniMax-M2.5
Result: ⚠️ launches, looks okay, but without animations, and nothing happens in app.
5h limit: 1% -> 4%
qwen3-coder-next
Result: ❌ fail, launches but only to show glitches.
5h limit: 4% -> 14%
kimi-k2.5
Result: ❌ fail. Opencode (or model) jumped in endless loop
glm-5
Result: ✅ Success (with 2 extra prompts to fix issues)
5h limit: 18% -> 21%
```
English









