GLM-4.5-Air

+ contribute

GLM-4.5-Air is an open-weights language model from Z.ai, released July 2025. Its published size is 106B parameters (12B active per token) and a 131k-token context window. Licence: mit.

62.7±0.8#167 of 719 LLMsfrom 33 benchmark results across 9 sources

$ how it compares

The LLMs ranked either side of GLM-4.5-Air, out of 828 with a score, then more from Z.ai.

#	model	score	where	RAM	released
193	o1-preview OpenAI	63±1	comparison only		2024-09
194	Gemini 2.5 Flash Preview (04-17) Google	63±1	comparison only		2025-04
195	GLM-4.5-Air Z.ai	63±1	api $0.42/M +local	~35GB	2025-07
196	Qwen3.5 9B Alibaba	63±1	api $0.11/M +local	~7.4GB	2026-03
197	Seed-OSS 36B Instruct ByteDance Seed	63±1	local	~23GB	2025-08
	more from Z.ai
33	GLM-5.3 Z.ai	76±1	api $2.15/M +local	~239GB	2026-08
52	GLM-5.3-Flash Z.ai	74±1	api $0.24/M +local	~103GB	2026-08
57	GLM-5.2 Z.ai	74±1	api $2.15/M +local	~121GB	2026-06

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #140 of 300 · 18% of its score · 9 tests

Beats Qwen3 235B-A22B Thinking 2507, o4-mini, o1 (2024-12-17) here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Large, GPT-4o (2024-05-13) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified lab	57.6% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 49/131	hard · avg 24%	▇ 3.6%	sharp · at its level · small run
τ-bench Retail lab	77.9% ✎	▁····▃·▂▇█▂▆ beats 24/28	medium · avg 45%	▇ 2.8%	at its level · small run
τ²-bench Telecom	28.1% ✎	▁·▅█▆▅▂▃▃▂▃▁ beats 22/89	hard · avg 15%	▇ 2.8%	sharp · at its level · noisy · small run
BrowseComp lab	21.3% ✎	▂·▄▁▃▃▆▄█▂▃▄ beats 7/42	very hard · avg 12%	▇ 2.1%	at its level · small run
τ-bench Airline lab	60.8% ✎	▁▁▁·▅▄▁█▇▁▁▇ beats 28/28	hard · avg 29%	▇ 1.9%	at its level · small run
Terminal-Bench (pre-2.0 release) lab	30.0% ✎	▃·▃···▃▃▅·▅█ beats 4/10	very hard · avg 15%	▇ 1.6%	at its level · small run
BFCL V4 (2025-11-03 archive, models absent from the December archive)	68.9% ✎	▃··▅▃▃·▅·▃·█ beats 8/10	hard · avg 34%	▇ 1.4%	at its level
Galileo Agent Leaderboard v2 (action completion)	44.0% ✎	▅▃·▃▃▅▅█▃█▅▅ beats 11/19	hard · avg 27%	▇ 1.2%	at its level

1 more test

test	result	vs the others in its run	difficulty	counts	why
BFCL v3 lab	76.4% ✎	▁▁▁▁▁▃▅█▆▅▄▃ beats 167/170	medium · avg 60%	▇ 0.7%	blunt · at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #161 of 458 · 7.4% of its score · 4 tests

Beats Gemini 2.5 Pro, Qwen3-Coder-Next (80B-A3B), Grok 4.1 Fast (Reasoning) here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 8B Reasoning 2512, Phi-4-reasoning 14B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Verified lab	57.6% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 49/131	hard · avg 24%	▇ 3.6%	sharp · at its level · small run
LiveCodeBench v6 (Aug 2024–May 2025)	61.5% ✎	▄▄·▄·····█·▄ beats 0/5	medium · avg 45%	▇ 2.4%	at its level
SciCode (level unspecified) lab	37.3% ✎	▁····▃▄▄▄█▆▂ beats 19/34	hard · avg 19%	▇ 1.5%	at its level · small run
LiveCodeBench 2024-07 to 2025-01 (version not stated) lab	70.7% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #135 of 393 * · 7.0% of its score · 2 tests

Beats o3-mini, EXAONE 4.5 33B, INTELLECT-3 here, though they rank higher overall. Trails LFM2.5 8B-A1B, LFM2.5 2.6B, Spark-X2.5 1.7B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1373.3 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 213/379	avg 1268	▇▇ 5.7%	blunt
IFEval prompt-level strict lab	86.3% ✎	▁▁··▂··▂▇█▃▃ beats 20/26	easy · avg 74%	▇ 1.3%	small run

math▁▂▃▅▆▅▄█▄▁▁▁ #93 of 398 · 21% of its score · 3 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, MiniMax-M3 here, though they rank higher overall. Trails Gemini 2.5 Flash here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2024 +1	84.6% ✎	█···███·█··█ beats 2/5	medium · avg 49%	▇▇▇▇▇ 11%	sharp · several runs
AIME 2025	82.0% ✎	█·█···███··█ beats 2/5	medium · avg 43%	▇▇ 6.0%	sharp
MATH-500 +1	97.8% ✎	▄······▄▄▄·█ beats 4/5	very easy · avg 91%	▇ 3.4%	maxed out · small run · several runs

web search▃▁▂▆▄█▃▃▁▁·▁ #80 of 80 * · 2.1% of its score · 1 test

Trails GPT-4.1 nano, GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp lab	21.3% ✎	▂·▄▁▃▃▆▄█▂▃▄ beats 7/42	very hard · avg 12%	▇ 2.1%	at its level · small run

other tests14 not in a task · 50% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	73.3% ✎	█··██·██···█ beats 1/5	hard · avg 42%	▇▇▇▇▇ 12%	sharp · at its level
Medmarks-V mean win rate (19 verifiable medical benchmarks, 71 configurations)	0.5410 ✎	▁▄▃▃▇▄█▃▅▁▁▁ beats 23/30	avg 0.423	▇▇▇▇ 8.8%	sharp
Humanity's Last Exam +1	13.3% ✎	▄····▄·█·▄·▄ beats 1/5	very hard · avg 4.3%	▇▇▇ 7.6%	several runs
MMLU-Pro +1	73.9% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 134/188	medium · avg 51%	▇▇ 5.8%	at its level · several runs
Medmarks-V weighted mean win rate (61 models, 71 configurations)	54.1% ✎	▂▃▅▂▆▃█▅█▅▃▆ beats 24/34	medium · avg 42%	▇▇ 4.3%	blunt · at its level
GPQA (subset unspecified) lab	75.0% ✎	▂▁▂▂▄▃▆▃▅█▄▃ beats 86/104	hard · avg 47%	▇▇ 3.9%	sharp · at its level · small run
MMLU lab	87.4% ✎	▁▁▁▁▁▂▃▅█▃▆▃ beats 169/183	medium · avg 72%	▇ 2.9%	small run
SimpleQA lab	14.5% ✎	▄▄▄▆█▆███·▂█ beats 16/32	very hard · avg 7.7%	▇ 1.9%	small run

6 more tests

test	result	vs the others in its run	difficulty	counts	why
MultiChallenge lab	42.5% ✎	▁·▁·▁▃▆▅▃▁█▅ beats 7/29	hard · avg 27%	▇ 1.5%	at its level · small run
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy	80.4% ✎	▃·▁▁▂▁▂▅█▇▄▁ beats 73/100	medium · avg 68%	▇ 0.6%	noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy	95.8% ✎	▁▁▁▁▁▁▁▂▄█▄▃ beats 144/146	easy · avg 83%	▇ 0.3%	noisy
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	98.1% ✎	▁▁▂▃▄▄█▆▃▃▂▁ beats 123/140	very easy · avg 94%	▇ 0.2%	blunt · maxed out · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	90.7% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 54/98	very easy · avg 90%	▇ 0.0%	blunt
SysBench (ISR) lab	77.4% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · $0.43/M · $0.2 in / $1.1 out per 1M tokens · 131k context · docs.z.ai ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run GLM-4.5-Air · needs ~64.1 GB · from huggingface.co/unsloth
chat in the terminal
llama cli -hf unsloth/GLM-4.5-Air-GGUF:UD-Q4_K_XL
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/GLM-4.5-Air-GGUF:UD-Q4_K_XL

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run GLM-4.5-Air · needs ~64.1 GB · from huggingface.co/unsloth
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/unsloth/GLM-4.5-Air-GGUF@UD-Q4_K_XL -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run GLM-4.5-Air · needs ~32.3 GB · from huggingface.co/unsloth
chat in the terminal
ollama run hf.co/unsloth/GLM-4.5-Air-GGUF:UD-Q2_K_XL

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run GLM-4.5-Air · needs ~60.1 GB, Mac only · from huggingface.co/lmstudio-community
chat in the terminal
mlx_lm.chat --model lmstudio-community/GLM-4.5-Air-MLX-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model lmstudio-community/GLM-4.5-Air-MLX-4bit

Weights and documentation: huggingface.co/zai-org ↗

$ how the score was worked out · GLM-4.5 Team (Zhipu AI & Tsin… 37%, Prime Intellect INTELLECT-3 m… 37%, Medmarks: A Comprehensive Ope… 13% and 6 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
GLM-4.5 Team (Zhipu AI & Tsinghua) — GLM-4.5 technical report · lab-reported	16	▇▇▇▇▇▇▇▇▇▇▇ 37%	CC BY 4.0
Prime Intellect INTELLECT-3 model card: Evaluation table (Environments Hub runs)	7	▇▇▇▇▇▇▇▇▇▇▇ 37%	MIT
Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 models	2	▇▇▇▇ 13%	CC BY 4.0
LMArena leaderboard dataset	1	▇▇ 6%	CC BY 4.0
Solar Open Technical Report (arXiv:2601.07022v1)	1	▇ 3%	CC-BY-4.0
Berkeley Function Calling Leaderboard result archive	1	▇ 1%	Apache-2.0
Galileo Agent Leaderboard (v1 and v2 results)	1	▇ 1%	Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard	2	▇ 1%	MIT
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	2	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 62.7

not counted (2)

GLM-4.5 Team (Zhipu AI & Tsinghua) — GLM-4.5 technical report	LiveCodeBench 2024-07 to 2025-01 (version not stated) [correct]	70.7%	fewer than three models have this test yet, or this source ran it for this model alone
GLM-4.5 Team (Zhipu AI & Tsinghua) — GLM-4.5 technical report	SysBench (ISR) [correct]	77.4%	fewer than three models have this test yet, or this source ran it for this model alone

$ published speed measurements · 1

Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.

AMD Ryzen AI Max+ 395 (Strix Halo), Radeon 8060S; pre-production Framework Desktop 128GB LPDDR5x-8000

UD-Q4_K_XL

23.35 tok/s

llama.cpp 7ad67ba9f (commit from system_info.json), backend vulkan_radv · llama-bench tg128 (128 generated tokens, empty context); backend variant 'llama.cpp-vulkan_radv'; flags: -fa 1; mmap disabled per the guide; value = bold best tg128 in this model's Benchmark Results table

Author's own sweep; README value 23.354522 rounded to 2 decimals. Kernel 6.16.0-1-mainline; system_info timestamp 2025-08-08T11:23:08.839151Z; CPU string 'AMD RYZEN AI MAX+ 395 w/ Radeon 8060S'. Pre-production hardware and BIOS per the README.

source ↗

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json