GLM-4.5V

GLM-4.5V places in the top three on 4 benchmarks.

Z.ai Released August 11th, 2025 Family GLM-V The lab's model card Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Z.ai Aug 2025

GLM-4.5V

Open weights Mid-size

135.6 #223 of 404

Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.

MMMU-Pro 65.2% #60 /90
ERQA 50% #28 /36
MMVU 68.7% #16 /26

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
HallusionBench	65.4%	#11 of 31	Lab model card
MMBench-CN v1.1	88.3%	#2 of 5	Lab model card
LMArena (Elo)	1,352	#175 of 335	Official leaderboard

Math and science

Benchmark	Score		Rank	Source
HiPhO	35.3%		#26 of 36	Official leaderboard

Coding

Benchmark	Score		Rank	Source
SArena	0.87		#2 of 17	Official leaderboard
Flame-React-Eval	82.5%		#3 of 4	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
WebVoyager	84.4%	#7 of 12	Lab model card
SnakeBench	21	#121 of 188	Official leaderboard
OSWorld	35.8%	#10 of 13	Lab model card
AndroidWorld	57%	#20 of 23	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
MMMU	75.4%	#20 of 136	Official leaderboard
SpaCE-10	51.6%	#6 of 36	Official leaderboard
CVBench	87.3%	#3 of 15	Lab model card
MathCanvas-Bench	35.6%	#4 of 16	Official leaderboard
MathVista	84.6%	#20 of 70	Lab model card
MM-HELIX	27%	#5 of 17	Official leaderboard
GSM8K-V	28.3%	#8 of 24	Official leaderboard
MME-CC	30.4%	#4 of 12	Official leaderboard
Ref-L4	89.5%	#1 of 3	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	91.3%	#10 of 29	Lab model card
Math-VR	49.6%	#10 of 26	Official leaderboard
Video Reality Test	63.6%	#2 of 5	Official leaderboard
BLINK	65.3%	#13 of 29	Lab model card
MuirBench	75.3%	#6 of 12	Lab model card
CompassArena Multimodal	1,089	#4 of 8	Official leaderboard
GeoBench	79.7%	#1 of 2	Lab model card
OCRBench	86.5%	#18 of 34	Lab model card
VideoMME v2 (with subtitles)	21.8%	#17 of 31	Official leaderboard
ZeroBench	23.4%	#17 of 31	Lab model card
Video-MME	80.7%	#28 of 51	Lab model card
MMStar	75.3%	#20 of 36	Lab model card
AI2D	88.1%	#29 of 52	Lab model card
LogicVista	62.4%	#8 of 14	Lab model card
MathVerse	72.1%	#7 of 12	Lab model card
MMBench-EN v1.1	88.2%	#14 of 23	Lab model card
MMVU	68.7%	#16 of 26	Lab model card
MMLongBench-Doc	44.7%	#11 of 17	Official leaderboard
MMMU-Pro	65.2%	#60 of 90	Lab model card
MATH-Vision (MathVision)	65.6%	#34 of 51	Lab model card
SpatialWorld Digital	14.5%	#6 of 9	Official leaderboard
All-Angles Bench	56.9%	#4 of 6	Lab model card
WebQuest	68.8%	#2 of 3	Lab model card
MVBench	73%	#16 of 23	Lab model card
ChartMuseum	40.6%	#15 of 21	Official leaderboard
TreeBench	50.1%	#5 of 7	Lab model card
Vision Arena	1,153	#99 of 136	Official leaderboard
ChartQA Pro	64%	#6 of 8	Lab model card
We-Math	68.8%	#10 of 13	Lab model card
ERQA	50%	#28 of 36	Lab model card
SpatialWorld Physical	4.00%	#7 of 9	Official leaderboard
DynaMath	53.9%	#18 of 23	Lab model card
Video-MMMU	72.4%	#29 of 36	Lab model card
OmniSpatial	51%	#5 of 6	Lab model card
LVBench	53.8%	#27 of 31	Lab model card
MotionBench	62.4%	#8 of 9	Lab model card
Math Kangaroo 2025 (MathArena)	67.6%	#21 of 22	Official leaderboard

Safety and security

Benchmark	Score		Rank	Source
nocot-bench (NCRI)	66.7		#113 of 185	Official leaderboard
BABE	20.8%		#6 of 6	Official leaderboard

Its lab's numbers put it at 135.6, number 223 of 404

The Self-Reported Index fits one score to the 28 capability results Z.ai reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 134.0 to 137.2. How we compute it.

Its lab's numbers next to 6 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
ChartMuseum	55.3% ChartMuseum	40.6% Leaderboard	+15 pts
MotionBench	62.4% MotionBench	62.4% Leaderboard, frames: 2fps	Same
MATH-Vision (MathVision)	65.6% MathVision	65.6% Leaderboard	Same
MMLongBench-Doc	44.7% MMLongBench-Doc	44.7% Leaderboard, VL; 12B activated (106B total)	Same
MMMU	75.4% MMMU (val)	75.4% Leaderboard, thinking	Same
MMMU-Pro	65.2% MMMU Pro	65.2% Leaderboard, thinking	Same

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

GLM-4.5V model card (Hugging Face) Aug 11, 2025

MMBench v1.1 88.2
MMBench v1.1 (CN) 88.3
MMStar 75.3
BLINK (val) 65.3
MUIRBENCH 75.3
HallusionBench 65.4
ZeroBench (sub) 23.4
GeoBench 79.7
MMMU (val) 75.4
MMMU Pro 65.2
MathVista 84.6
MathVision 65.6
MathVerse 72.1
DynaMath 53.9
LogicVista 62.4
AI2D 88.1
WeMath 68.8
MMLongBench-Doc 44.7
OCRBench 86.5
ChartQAPro 64
ChartMuseum 55.3
RefCOCO-avg (val) 91.3
TreeBench 50.1
Ref-L4-test 89.5
OmniSpatial 51
CV-Bench 87.3
ERQA 50
All-Angles Bench 56.9
OSWorld 35.8
AndroidWorld 57
WebVoyagerSom 84.4
Mean of Webquest-SingleQA 76.9 / Webquest-MultiQA 60.6 68.8
Design2Code 82.2
Flame-React-Eval 82.5
VideoMME (w/ sub) 80.7
MMVU 68.7
VideoMMMU 72.4
LVBench 53.8
MotionBench 62.4
MVBench 73

Its system card

Every result in the longer documents the lab published with it.

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning Aug 13, 2025