‹ PublicAI Index
The LLM benchmark aggregator.
GLM-5.2
Z.ai · 753B · open weights
Strongest in Human preference (#30 of 354), weakest in Safety (#257 of 329). Above par in 25 of 33 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Fable 5 and ahead of Claude 3.7 Sonnet and MiniMax M3.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Human preference63.9−4.9#30/3541/1
Human preference63.9−4.9#30/3541/1
Agents55.6−10.8#51/2982/6
Knowledge work59−11.7#53/2082/2
Tool use60−14—/81✱0/1
Workflow automation54.2−11.8—/0✱✱
Coding53.4−12.8#52/1792/6
Agentic coding53.5−13.4#53/1732/5
Code generation62−3.3—/20✱0/1
Repository Q&A54.8−11.9—/0✱✱
Professional53.1−9.3#59/1941/2
Finance55.3−7.1#44/1621/1
Legal52.6−13.3#71/1731/1
Medical51.3−15.2#77/1491/1
Knowledge52.9−9.6#69/1971/3
Academic knowledge56.9−7#38/1231/1
Factuality54.5−6.3—/0✱✱
Reasoning51.8−13.6#82/1954/5
Science56.6−6.9#47/1221/1
Reasoning50−19#66/1472/3
Mathematics50.8−13#81/1521/2
Expert reasoning63.8−2.6—/0✱✱
Core abilities47.8−17.5#118/1992/4
Language47.5−19.1#40/621/1
Instruction following39.9−33.9#51/621/1
General intelligence53.6−15.8#60/1781/3
Long context60.2−4.4—/0✱✱
Safety46.5−14.9#257/3291/3
Secure code59.5−6.7#69/2701/1
Toxicity avoidance55.1−3.9#102/2681/1
Fairness44.5−25.5#198/2921/2
Harm refusal44.8−16.7#243/2921/2
Jailbreak resistance30.6−34.4#251/2681/1
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for GLM-5.2, left for the other.
§ 3 · Sources
Where the numbers come from
11 publications, 55 figures. Every one links to the page it was read from.
LMArena Text 1475
GDPval-AA 1381
AA-Briefcase 1226
ARC-AGI-2 22.8%
LiveBench 1.0 73.2LiveBench 1.0 · Reasoning 78.6LiveBench 1.0 · Coding 79.7LiveBench 1.0 · Agentic Coding 51.8LiveBench 1.0 · Mathematics 89.8LiveBench 1.0 · Data Analysis 73.7LiveBench 1.0 · Language 76.2LiveBench 1.0 · Instruction Following 62.3
Kagi LLM Benchmark 60%
SimpleBench 58.8%
Vals · Legal Research Bench 31.25%Vals · LegalBench 84.07%Vals · Finance Agent 49.7%Vals · CorpFin 66.12%Vals · TaxEval 73.34%Vals · MedCode 40.77%Vals · MedScribe 83.53%Vals · SWE-bench Verified 82.8% (Mini-SWE-agent)Vals · Vibe Code Bench 63.96% (OpenHands)Vals · Code Migration 37.87%Vals · GPQA Diamond 85.61%Vals · MMLU Pro 86.71%
Enkrypt · Jailbreak risk 29.1%Enkrypt · Harmful content risk 9.4%Enkrypt · CBRN risk 50.7%Enkrypt · Toxicity risk 2.7%Enkrypt · Bias risk 86.3%Enkrypt · Insecure code risk 13.8%
GDPVal-AA 1498tau3-Banking 34.6%Toolathlon Verified 59.9%Automation Bench Public 26.2%Apex-Agents (pass@1) 26.9%MCPMark 72.4%WildClawBench 55%Terminal-Bench 2.1 77.9%SciCode 50.5%SWE-Atlas-QnA 46.4%SWE Bench Pro 46.7%Humanity's Last Exam (without tools) 41.1%GPQA Diamond 89.5%CritPt 20.9%AA-LCR 76.7%AA-Omniscience Accuracy 24%AA-Omniscience Non-Hallucination 74%
GDPVal-AA (Elo) 1498Toolathlon Verified 59.9%Terminal-Bench 2.1 77.9%SWE-bench Pro (strict) 46.7%MCPMark 72.4%SWE-Atlas-QnA (strict) 46.4%
Badge
[](https://publicai.io/model-index/m/glm-5-2)