Evaluation overview

Overall rank

#7

94.6 points · 197 models

Open-weight rank

#2

96.9 points · 118 models

Data coverage

14 items

1 source · updated through 2026-08-26

Domain performance

Coding#12 / 177

88.8

Research#8 / 191

96

Agents#7 / 143

89.8

Vision#9 / 78

86.1

Agents · GLM-5.3-Flash Official Release

78.4%

1/ 23

Reasoning · GLM-5.3-Flash Official Release

1,773pts

2/ 27

Multimodal · GLM-5.3-Flash Official Release

89.4%

2/ 12

Coding · GLM-5.3-Flash Official Release

56.3%

7/ 37

Agents · GLM-5.3-Flash Official Release

48.8%

2/ 24

Multimodal · GLM-5.3-Flash Official Release

78%

1/ 5

Context (input / output): 1.05M / 131K
Parameters: 320B / 18B active
Release date: 2026-08
License: MIT
Input modalities: Text · Image · Video · file
Output modalities: Text
Layers: 45
Architecture: MoE, sparse + linear attention, mHC, IndexPool
Total parameters: 320B
Exact release date: 2026-08-26
Availability: GLM Coding Plan and API
API model ID: glm-5.3-flash
Reasoning mode: Thinking only (supports low / high / max)
Weight status: Open weights released (MIT)
Evaluation source: Official 14-row comparison table with 70 scores; base-model table excluded
Pretraining data: 30T-token multimodal corpus
Default reasoning effort: max (supports low / high / max)