Model

ChatGLM3-6B

OPEN

Zhipu AI / THUDM (Tsinghua) · ChatGLM family · released Oct 27, 2023

Third-gen 6B model adding native function calling, code interpreter and agent tool use; strongest sub-10B base model on its release benchmarks.

ReasoningCodingVisionFunction callingTool useAgentic

367.9

Elo · rank #340

Parameters

6.2B

Active params

6.2B (dense)

Context

8K tokens

Architecture

Dense Transformer (GLM, Multi-Query Attention, SwiGLU, 28 layers)

License

Apache-2.0 (code); ChatGLM3-6B Model License for weights (free for research; free commercial use after registration)

Languages

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AGIEvalReasoning53.7%#12

best: OLMo 3-Think 32B · 88.2%

BIG-Bench HardReasoning66.1%#69

best: ERNIE 4.5 300B-A47B · 94.3%

C-EvalKnowledge69.0%#42

best: Qwen3.6-Plus · 93.3%

CMMLUKnowledge67.5%#21

best: Doubao-1.5-Pro · 90.9%

DROPReasoning70.9#44

best: Hunyuan-T1 · 93.1

GPQA DiamondReasoning26.8%#270

best: GPT-5.6 · 94.6%

GSM8KMath72.3%#97

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning79.7%#85

best: Claude 3 Opus · 95.4%

HumanEvalCoding58.5%#105

best: Claude Opus 4.5 · 99.4%

MATH-500Math25.7%#171

best: GPT-5 · 99.4%

MBPPCoding52.4%#69

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLUKnowledge61.4%#199

best: OpenAI o3 · 92.9%

PIQAReasoning80.1%#42

best: GPT-4o mini · 93.1%

Run it locally

VRAM @ Q4

6 GB

VRAM @ FP16

13 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

INT4 · INT8

ChatGLM family

Elo progression across releases

ChatGLM-6BMar 2023-456.9 ChatGLM2-6BJun 2023-219.2 ChatGLM3-6BOct 2023367.9

Resources

HFzai-org/chatglm3-6b GHzai-org/ChatGLM3

API price weights · each benchmark row carries its own source badge (see methodology)