Model
Model explorer

GLM-130B

OPEN
Zhipu AI / THUDM (Tsinghua) · GLM family · released Aug 1, 2022

GPT-3-scale 130B bilingual (EN/ZH) dense model; open code, weights and training logs (ICLR 2023).

ReasoningCodingVisionFunction callingTool useAgentic
-67.8
Elo · rank #409
Parameters
130B
Active params
130B (dense)
Context
2K tokens
Architecture
Dense bidirectional Transformer (GLM objective)
License
Custom GLM-130B Model License (non-commercial research; code Apache-2.0)
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
BIG-benchReasoning13.3%#7
best: Chinchilla · 65.1%
LAMBADAReasoning80.2%#3
best: GPT-3 175B · 86.4%
MMLUKnowledge44.8%#249
best: OpenAI o3 · 92.9%
Run it locally
VRAM @ Q4
96 GB
VRAM @ FP16
—
Fits on (Q4)
M3 Ultra 512GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
INT4 · INT8
GLM family
Elo progression across releases
GLM-130BAug 2022-67.8
Resources
GHzai-org/GLM-130BDOCSkeg.cs.tsinghua.edu.cn/glm-130b/posts/glm-130b/
API price weights · each benchmark row carries its own source badge (see methodology)