ModelRadar
/

GLM-5.2

Z.ai (Zhipu)🇨🇳 CNFrontier· Jun 20, 2026

TextReasoningopen weightsAPI✎ Editorial

GLM-5.2 is an open-source model from Z.AI for long-horizon tasks, with a stable 1-million-token context and strong agentic coding. It combines code generation with tool use and supports adjustable thinking-effort levels. It scores 62.1% on SWE-bench Pro, sits close to the frontier on AIME and GPQA-Diamond, and improves markedly over GLM-5.1. The MoE architecture with IndexShare cuts compute per token by a factor of 2.9.

Editorial

Context
1M tokens
Params
753B MoE
License
GLM Community License
Modalities
text, code → text, code

2 variants

Standard753B MoE1MHF · OR
FP8HF

Sources

HuggingFaceOpenRouterPaperWebsite
Deutsch

GLM-5.2 ist ein Open-Source-Sprachmodell von Z.AI mit stabilem Kontextfenster von einer Million Tokens sowie agiler Coding- und Tool-Nutzung. Es unterstuetzt Text, Bild- und Audiokontext sowie Mehrsprachigkeit in Englisch, Chinesisch und Arabisch. Bei SWE-bench Pro erzielte es 62,1 Prozent, bei AIME 2026 99,2 Prozent Loesungsrate und im Terminal-Bench 81,0 Prozent, wobei es sich deutlich gegenueber GLM-5.1 verbessert hat. Das MoE-Design mit IndexShare reduziert den Rechenaufwand erheblich.

Benchmarks

BenchmarkDomainValueEvidence
HMMTvendor numbermath94.4 % correct61
AIMEvendor numbermath99.2 % correct53
GPQA Diamondvendor numberreasoning91.2 % correct50

Vendor-reported numbers are marked and count with factor 0.7 in rankings. The evidence score says how much a benchmark still tells you today.

Provenance

Detected
Jun 20, 2026
First source
kuratiert
Description
Editorial
Editorially reviewed
yes