Skip to content
Updated minutes ago
Zhipu

ChatGLM3 6B

Tsinghua University · dense · 6B parameters · 131,072 context

Quality
50.0

Parameters

6B

Context Window

128K tokens

Architecture

Dense

Best GPU

A10G

Intelligence Brief

ChatGLM3 6B is a 6B parameter DENSE model from Tsinghua University, featuring Grouped Query Attention (GQA) with 28 layers and 4,096 hidden dimensions. With a 131,072 token context window, it supports tools, code, math, multilingual. For self-hosted inference, A10G delivers optimal throughput at $285/month.

Architecture Details

TypeDENSE
Total Parameters6B
Active Parameters6B
Layers28
Hidden Dimension4,096
Attention Heads32
KV Heads2
Head Dimension128
Vocab Size65,024

Memory Requirements

BF16 Weights

12.0 GB

FP8 Weights

6.0 GB

INT4 Weights

3.0 GB

KV-Cache per Token28672 bytes
Activation Estimate0.80 GB

GPU Compatibility Matrix

ChatGLM3 6B is compatible with 95% of GPU configurations across 41 GPUs at 3 precision levels.

BF16 (Full)
FP8 (Half)
INT4 (Quarter)
Blackwell(7 GPUs)
B200 NVL (pair)360GB
B300288GB
B100 SXM192GB
GB200 NVL72 (per GPU)192GB
Hopper(7 GPUs)
H100 NVL 94GB (per GPU pair)188GB
H200 SXM141GB
H2096GB
GH20096GB
Ada Lovelace(11 GPUs)
L40S48GB
L4048GB
RTX 6000 Ada48GB
L2048GB
Ampere(16 GPUs)
A100 80GB SXM80GB
A100 80GB PCIe80GB
A1664GB
RTX A600048GB
Legend:No fitVery tightTightModerateGoodExcellent

GPU Recommendations

A10Goptimal

BF16 · 1 GPU · vllm

100/100

score

Throughput

270.0 tok/s

Latency (ITL)

3.7ms

Est. TTFT

1ms

Cost/Month

$285

Cost/M Tokens

$0.40

Use this config →
A30optimal

BF16 · 1 GPU · vllm

100/100

score

Throughput

419.8 tok/s

Latency (ITL)

2.4ms

Est. TTFT

0ms

Cost/Month

$332

Cost/M Tokens

$0.30

Use this config →
RTX 4090optimal

BF16 · 1 GPU · vllm

100/100

score

Throughput

453.6 tok/s

Latency (ITL)

2.2ms

Est. TTFT

0ms

Cost/Month

$370

Cost/M Tokens

$0.31

Use this config →

Deployment Options

API

API Deployment

No API pricing available

Self-Hosted

Single GPU

A10G

$285/mo

Min VRAM: 6 GB

Scale

Multi-GPU

RTX 3080 x2

518.6 tok/s

TP· $266/mo

API Pricing Comparison

No API pricing data available for this model.

Performance Estimates

Throughput by GPU

A10G
270.0 tok/s
A30
419.8 tok/s
RTX 4090
453.6 tok/s

VRAM Breakdown (A10G, BF16)

Weights
Act
Weights 12.0 GBKV-Cache 0.5 GBActivations 6.4 GBOverhead 1.0 GB

Precision Impact

bf16

12.0 GB

weights/GPU

~270.0 tok/s

fp8

6.0 GB

weights/GPU

int4

3.0 GB

weights/GPU

Capabilities

Features

Tool Use Vision Code Math Reasoning Multilingual Structured Output

Supported Frameworks

vllmsglangtgiollama

Supported Precisions

BF16 (default)FP8INT4

Where to Deploy ChatGLM3 6B

Similar Models

Smaller contextCompare →

StarCoder2 7B

6.73B params · dense

Quality: 35

from $0.15/M

Smaller context, Lower qualityCompare →

Frequently Asked Questions

How much VRAM does ChatGLM3 6B need for inference?

ChatGLM3 6B requires approximately 12.0 GB of VRAM at BF16 precision, 6.0 GB at FP8, or 3.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (28672 bytes per token) and activations (~0.80 GB).

What is the best GPU for ChatGLM3 6B?

The top recommended GPU for ChatGLM3 6B is the A10G using BF16 precision. It achieves approximately 270.0 tokens/sec at an estimated cost of $285/month ($0.40/M tokens). Score: 100/100.

How much does ChatGLM3 6B inference cost?

ChatGLM3 6B inference costs vary by provider and GPU setup. Use our calculator for detailed cost estimates across all providers.