BREAKING
Sonnet 5 Trails GLM-5.2 in 3D Test
The Monica's apt test
1
Read floor plan
↓
2
Generate threejs code
↓
3
Door placement off
↓
4
Chair orientation off
Open weight vs proprietary
GLM-5.2
Open-weight
●
MIT license
●
Self-host and fine-tune
●
From Zhipu AI
Claude Sonnet 5
Proprietary
●
Anthropic
●
No self-hosting
●
Closed weights
Output cost per 1M tokens
GLM-5.2 out
6
Sonnet 5 out
10
0
%
GLM-5.2 F1
0
%
Claude Code F1
Open models close the gap
AI NEWS BLITZ
Claude Sonnet 5 was judged worse than the open-weight GLM-5.2 in a 3D recreation test.