zai-org/GLM-5.2
Capability leader
SADB V1 · Reference snapshot
Seven views of the same 72-question bank.
These results are a provider-path snapshot collected with the frozen Chinese V1 bank. They are useful for understanding strengths and gaps, not for making a universal claim about a model.
Model ranking
Capability points and request stability
Collected 2026-08-31, Chinese bank, five identical stability samples. A score is shown as raw points followed by completion of the 72-question bank.
55 (76.39%)100% stable
moonshotai/Kimi-K2.7-Code
Strong all-rounder
deepseek-ai/DeepSeek-V4-Pro
Reliable depth
Qwen/Qwen3.5-397B-A17B
deepseek-ai/DeepSeek-V4-Flash
nex-agi/Nex-N2-Pro
meituan-longcat/LongCat-2.0
Seven-model comparison
A vertical view keeps the gaps visible.
Chinese and English completion points are shown side by side for all seven tested models.
System and hardware diagnostics14 questionsGLM 11 · Kimi 10 · Pro 9 · Flash 3 · Nex 2
Desktop operations and task planning12 questionsGLM 9 · Kimi 7 · Pro 8 · Flash 4 · Nex 2
File, document and data processing10 questionsGLM 8 · Kimi 7 · Pro 6 · Flash 5 · Nex 4
Software configuration and troubleshooting10 questionsGLM 5 · Kimi 7 · Pro 7 · Flash 2 · Nex 2
Programming, scripting and automation10 questionsGLM 7 · Kimi 7 · Pro 7 · Flash 3 · Nex 3
Security, permissions and privacy8 questionsGLM 8 · Kimi 8 · Pro 8 · Flash 5 · Nex 5
General logic and mathematical reasoning8 questionsGLM 7 · Kimi 7 · Pro 7 · Flash 3 · Nex 3