Model

GLM-4.5 (355B-A32B)

OPEN

Zhipu AI / Z.ai (Tsinghua) · GLM-4.5 family · released Jul 28, 2025

MIT-licensed agentic/reasoning/coding (ARC) MoE flagship; ranked 3rd overall and 2nd on agentic benchmarks among evaluated models at launch.

ReasoningCodingVisionFunction callingTool useAgentic

1759.0

Elo · rank #82

Parameters

355B

Active params

32B (MoE)

Context

128K tokens

Architecture

Mixture-of-Experts, 355B total / 32B active, hybrid thinking/non-thinking response modes

License

MIT

Languages

—

API price (in/out)

$0.6 / $2.2

Modalities

text

Benchmark results

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

AIMEMath91.0%#44

best: GPT-5.2 · 100.0%

BFCL v3Agents77.8%#2

best: Hunyuan-A13B · 78.3%

BrowseCompAgents26.4%#40

best: Kimi K3 · 91.2%

GPQA DiamondReasoning79.1%#85

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning14.4%#76

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning44.0%#41

best: MiniMax M3 · 83.0%

IFEvalReasoning86.1%#50

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding72.9%#62

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math98.2%#4

best: GPT-5 · 99.4%

MMLU-ProKnowledge84.6%#29

best: Claude Fable 5 · 91.5%

MMLUKnowledge90.0%#10

best: OpenAI o3 · 92.9%

SimpleQAKnowledge26.4%#18

best: GPT-4.5 · 62.5%

SWE-bench VerifiedCoding64.2%#69

best: Claude Fable 5 · 95.0%

tau-benchAgents70.1%#7

best: Claude Opus 4.1 · 82.4%

Terminal-BenchCoding37.5%#9

best: Claude Opus 4.5 (High) · 59.3%

Run it locally

VRAM @ Q4

180 GB

VRAM @ FP16

710 GB

Fits on (Q4)

M3 Ultra 512GB

355B MoE needs multi-GPU (8x H100/H200-class) even at FP8; does not fit a single RTX 4090.

Quantizations

GGUF · AWQ · FP8

GLM-4.5 family

Elo progression across releases

GLM-4.5 (355B-A32B)Jul 20251759.0 GLM-4.5-Air (106B-A12B)Jul 20251617.7

Resources

HFzai-org/GLM-4.5 GHzai-org/GLM-4.5 DOCSdocs.z.ai/guides/overview/pricing APIAvailable via Zhipu AI / Z.ai (Tsinghua) API

API price $0.6/$2.2 · each benchmark row carries its own source badge (see methodology)