‹ PublicAI Index
The LLM benchmark aggregator.
GLM-4.6
Z.ai · 357B · open weights
Strongest in Tool use (#4 of 81), weakest in Harm refusal (#255 of 292). Above par in 14 of 25 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Fable 5 and ahead of GLM-4.5 and Gemini 3.5 Flash-Lite.
§ 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.
Agents57.1−9.3#34/2982/6
Tool use74leads#4/811/1
Knowledge work48.9−21.8#105/2081/2
Reasoning53−12.4#66/1951/5
Mathematics57−6.8#31/1521/2
Science48.9−14.6#80/1221/1
Professional50.3−12.1#92/1941/2
Medical54.7−11.8#46/1491/1
Legal51−14.9#84/1731/1
Finance47.5−14.9#112/1621/1
Human preference59−9.8#104/3541/1
Human preference59−9.8#104/3541/1
Knowledge50.5−12#106/1971/3
Academic knowledge51.2−12.7#75/1231/1
Core abilities47.6−17.7#120/1991/4
General intelligence46.5−22.9#114/1781/3
Coding42.1−24.1#154/1791/6
Agentic coding41.4−25.5#152/1731/5
Safety47.1−14.3#249/3292/3
Factual grounding51.1−19.6#52/1041/1
Secure code55.8−10.4#110/2701/1
Toxicity avoidance52.3−6.7#165/2681/1
Jailbreak resistance40.6−24.4#211/2681/1
Fairness43.9−26.1#217/2921/2
Harm refusal43.8−17.7#255/2921/2
§ 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-4.6, left for the other.
§ 3 · Sources
Where the numbers come from
7 publications, 19 figures. Every one links to the page it was read from.
LMArena Text 1424
GDPval-AA 764
BFCL v4 72.38%
Kagi LLM Benchmark 45.7%
Vals · LegalBench 79.61%Vals · CorpFin 56.84%Vals · TaxEval 66.23%Vals · MedQA 92.22%Vals · Vibe Code Bench 3.09% (OpenHands)Vals · GPQA Diamond 74.5%Vals · MMLU Pro 82.2%Vals · AIME 92.71%
Enkrypt · Jailbreak risk 20.7%Enkrypt · Harmful content risk 45%Enkrypt · CBRN risk 25.3%Enkrypt · Toxicity risk 4.6%Enkrypt · Bias risk 87.3%Enkrypt · Insecure code risk 21.3%
Vectara · Factual consistency 90.5%
Badge
[](https://publicai.io/model-index/m/glm-4-6)