Request a Pilot
‹ PublicAI Index

The LLM benchmark aggregator.

GLM-4.5 Air

Z.ai · 111B · open weights

Strongest in Safe-prompt compliance (#5 of 82), weakest in Harm refusal (#162 of 292). Above par in 6 of 9 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 4.8 and GPT-5.5 and ahead of Llama 4 Maverick Instruct and Claude Sonnet 4.6.

§ 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.

Safety52.6#111/329
Safe-prompt compliance59.9#5/82
Fairness57.8#46/292
Factual grounding51.6#46/104
Harm refusal49.2#162/292
Core abilities46.7#129/199
General intelligence45.2#122/178
Human preference54.1#159/354
Human preference54.1#159/354

§ 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-4.5 Air, left for the other.

Jamba Mini 1.7won 6 · lost 0
Jamba Large 1.7won 6 · lost 0
DeepSeek LLM 67B Chatwon 6 · lost 0
DBRX Instructwon 6 · lost 0
Mixtral 8x7B Instructwon 6 · lost 0
GPT-3.5 Turbowon 6 · lost 0
Mixtral 8x22B Instructwon 6 · lost 0
Command Rwon 6 · lost 0
Llama 3.1 70B Instructwon 6 · lost 0
Mercury 2won 6 · lost 0
Magistral Mediumwon 7 · lost 0
Command Awon 7 · lost 0
Llama 4 Scout Instructwon 8 · lost 1
Claude 3 Haikuwon 7 · lost 1
Llama 3 70B Instructwon 7 · lost 1
Gemma 3 4B ITwon 7 · lost 1
Gemini 2.0 Flashwon 7 · lost 1
Mistral Small 3 24B Instructwon 6 · lost 1
Command R+won 6 · lost 1
Llama 3.1 8B Instructwon 5 · lost 1
Llama 3 8B Instructwon 5 · lost 1
Qwen1.5 72B Chatwon 5 · lost 1
OLMo 2 32B Instructwon 5 · lost 1
Gemini 1.5 Flashwon 5 · lost 1
Llama 3.1 405B Instructwon 5 · lost 1
Gemini 1.5 Prowon 5 · lost 1
Gemini 2.0 Flash-Litewon 5 · lost 1
GPT-4o miniwon 6 · lost 2
GPT-4.1 nanowon 6 · lost 2
Gemma 3 27B ITwon 6 · lost 2
GPT-4owon 5 · lost 2
Kimi K2.6won 5 · lost 2
Aya Expanse 8Bwon 4 · lost 2
Qwen2 72B Instructwon 4 · lost 2
Aya Expanse 32Bwon 4 · lost 2
Claude 3 Sonnetwon 4 · lost 2
Llama 3.3 70B Instructwon 4 · lost 2
GPT-4 Turbowon 4 · lost 2
O1 Miniwon 4 · lost 2
Gemma 3 12B ITwon 4 · lost 2
O3 Miniwon 4 · lost 2
Grok 4.1 Fastwon 4 · lost 2
Grok 4.3won 4 · lost 2
GPT OSS 120Bwon 6 · lost 3
GPT OSS 20Bwon 5 · lost 3
Gemma 4 31Bwon 5 · lost 3
Mistral Large 4won 4 · lost 3
NVIDIA Nemotron 3 Ultra 550B A55Bwon 4 · lost 3
MiniMax M3won 4 · lost 3
Claude Sonnet 4.6won 4 · lost 3
Llama 4 Maverick Instructwon 5 · lost 4
Amazon Nova Microwon 3 · lost 3
Phi 4won 3 · lost 3
Amazon Nova Litewon 3 · lost 3
Granite 4.0 H Smallwon 3 · lost 3
Amazon Nova Prowon 3 · lost 3
Qwen2.5 72B Instructwon 3 · lost 3
Claude 3 Opuswon 3 · lost 3
Gemma 4 26B A4Bwon 3 · lost 3
Qwen3.7 Maxwon 3 · lost 3
Gemini 3.5 Flashwon 3 · lost 3
Gemini 3.6 Flashwon 3 · lost 3
Grok 3 Miniwon 4 · lost 4
Qwen3 Next 80B A3B Thinkingwon 4 · lost 4
Grok 4won 4 · lost 4
GLM-4.6won 4 · lost 4
Kimi K2.5won 4 · lost 4
GPT-5 nanowon 4 · lost 5
Gemini 2.5 Flash-Litewon 4 · lost 5
DeepSeek V3won 4 · lost 5
Grok 3won 4 · lost 5
Kimi K2won 4 · lost 5
DeepSeek R1won 4 · lost 5
GLM-4.5won 3 · lost 4
Claude Haiku 4.5won 3 · lost 4
DeepSeek V4 Flashwon 3 · lost 4
MiMo V2.6 Flashwon 3 · lost 4
GLM-5.3 Flashwon 3 · lost 4
GLM-5.2won 3 · lost 4
Gemini 3.8 Flashwon 3 · lost 4
DeepSeek V3.1won 3 · lost 5
DeepSeek V3.2won 3 · lost 5
Claude Opus 4.6won 3 · lost 5
Qwen3 32Bwon 2 · lost 4
Claude 3.7 Sonnetwon 2 · lost 4
Qwen3.5 35B A3Bwon 2 · lost 4
GPT-5.4 nanowon 2 · lost 4
O1won 2 · lost 4
Grok 4 Fastwon 2 · lost 4
Mistral Large 3won 2 · lost 4
GPT-4.5won 2 · lost 4
GPT-5.4 miniwon 2 · lost 4
GLM-5won 2 · lost 4
Claude Opus 4.5won 2 · lost 4
Gemini 4 Argonwon 2 · lost 4
O4 Miniwon 3 · lost 6
Gemini 2.5 Flashwon 3 · lost 6
Gemini 3 Prowon 3 · lost 6
MiniMax M2won 2 · lost 5
Inklingwon 2 · lost 5
GPT-5.6 Lunawon 2 · lost 5
GPT-5.1won 2 · lost 5
GPT-6 Solwon 2 · lost 5
MiMo V2.6 Prowon 2 · lost 5
GPT-4.1 miniwon 2 · lost 6
O3won 2 · lost 6
DeepSeek V4 Prowon 2 · lost 6
GPT-5.6 Solwon 2 · lost 6
Claude Opus 4.7won 2 · lost 6
GPT-5 miniwon 2 · lost 7
Claude Sonnet 4won 2 · lost 7
GPT-4.1won 2 · lost 7
Claude Opus 4won 2 · lost 7
GPT-5won 2 · lost 7
Gemini 2.5 Prowon 2 · lost 7
Claude Sonnet 4.5won 2 · lost 7
Claude 3.5 Sonnetwon 1 · lost 5
MiniMax M2.5won 1 · lost 5
Qwen3.5 27Bwon 1 · lost 5
MiniMax M2.7won 1 · lost 5
Qwen3.5 122B A10Bwon 1 · lost 5
Gemini 3.1 Flash-Litewon 1 · lost 5
Claude Sonnet 5won 1 · lost 5
GPT-6 Astrawon 1 · lost 5
Qwen3 235B A22B Instructwon 1 · lost 6
GPT-6 Lunawon 1 · lost 6
Grok 4.5won 1 · lost 6
Qwen3.8 Maxwon 1 · lost 6
GPT-5.2won 1 · lost 7
GPT-5.4won 1 · lost 7
Claude Opus 5.5won 1 · lost 7
DeepSeek V4.1 Flashwon 0 · lost 6
Claude Opus 5won 0 · lost 6
Muse Spark 1.1won 0 · lost 6
Qwen3.5 397B A17Bwon 0 · lost 7
Grok 4.7won 0 · lost 7
Grok 4.6won 0 · lost 7
GPT-5.6 Terrawon 0 · lost 7
GPT-5.5won 0 · lost 7
Claude Opus 4.8won 0 · lost 7

§ 3 · Sources

Where the numbers come from

4 publications, 8 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-10
LMArena Text 1373
Kagi LLM Benchmark ↗1 measureread 2026-11-10
Kagi LLM Benchmark 43%
HELM Safety ↗5 measuresread 2026-11-10
HELM Safety · HarmBench 56.1%HELM Safety · SimpleSafetyTests 99%HELM Safety · Anthropic Red Team 99%HELM Safety · BBQ 97.8%HELM Safety · XSTest 98.6%
Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10
Vectara · Factual consistency 90.7%

Badge

PublicAI Index badge for GLM-4.5 Air[![PublicAI Index](https://publicai.io/model-index/badge?model=glm-4-5-air)](https://publicai.io/model-index/m/glm-4-5-air)