Qwen3-VL-235B-A22B

Qwen3-VL-235B-A22B ranks first on 5 of the 96 benchmarks with five or more models scored.

Qwen Released September 23rd, 2025 Family Qwen3-VL The lab's model card Compare with others

105

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Qwen Sep 2025

Qwen3-VL-235B-A22B

Open weights Flagship

140.6 #173 of 404

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

MMMU-Pro 69.3% #52 /90
SimpleVQA 61.3% #19 /30
ERQA 52.5% #26 /36

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
MMLU	90.6%	#8 of 131	Lab model card
WritingBench	86.7	#2 of 14	Lab model card
Creative Writing v3	85.7	#2 of 12	Lab model card
MMBench-CN v1.1	88.6%	#1 of 5	Lab model card
MMLU-Redux	93.7%	#10 of 45	Lab model card
Multi-IF	79.1%	#7 of 28	Lab model card
Arena-Hard v2.0	74.8%	#6 of 24	Lab model card
MMLU-Pro	83.8%	#43 of 167	Lab model card
SimpleQA	44.4%	#16 of 60	Lab model card
HallusionBench	66.7%	#10 of 31	Lab model card
IFEval	88.2%	#52 of 135	Lab model card
INCLUDE	80%	#12 of 31	Lab model card
LMArena (Elo)	1,395	#137 of 335	Official leaderboard
MMLU-ProX	80.6%	#10 of 23	Lab model card
SuperGPQA	64.3%	#16 of 35	Lab model card
SimpleVQA	61.3%	#19 of 30	Lab model card
Humanity's Last Exam	13.6%	#127 of 154	Lab model card
SIFo	77.3%	#1 of 1	Lab model card

Math and science

Benchmark	Score	Rank	Source
ZebraLogic	97.3%	#1 of 23	Lab model card
HiPhO	62.8%	#8 of 36	Official leaderboard
AIME 2025	89.7%	#51 of 143	Lab model card
GPQA Diamond	77.1%	#136 of 302	Lab model card
PolyMath	57.8%	#11 of 24	Lab model card
HMMT 2025	77.4%	#40 of 54	Lab model card
SGI-Bench	28.3%	#12 of 15	Official leaderboard

Coding

Benchmark	Score	Rank	Source
Design2Code	93.4	#1 of 2	Lab model card
LiveCodeBench v6	70.1%	#55 of 90	Lab model card
Codeforces rating	1,964	#29 of 47	Lab model card
OJBench	27.5%	#13 of 16	Lab model card
ChartMimic	78.4	#1 of 1	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
ScreenSpot	95.4%	#1 of 11	Lab model card
Berkeley Function Calling Leaderboard (BFCL)	71.9%	#12 of 41	Lab model card
OSWorld-G	68.3%	#2 of 5	Lab model card
SnakeBench	25	#85 of 188	Official leaderboard
OSWorld	38.1%	#7 of 13	Lab model card
OSWorld-MCP	39.5%	#5 of 9	Official leaderboard
GameHorizon	65.2	#22 of 37	Official leaderboard
ScreenSpot-Pro	61.8%	#21 of 32	Lab model card
AndroidWorld	62%	#17 of 23	Lab model card
Windows Agent Arena	32.1%	#4 of 5	Lab model card
MobileWorld	9.49%	#16 of 18	Official leaderboard
GameWorld	31.4%	#20 of 22	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
Math-VR	66.8%	#1 of 26	Official leaderboard
DocVQA	96.5%	#2 of 36	Lab model card
MMMU	78.7%	#10 of 136	Official leaderboard
ChartQA	90.3%	#4 of 37	Lab model card
OVO-S-Bench	53.6%	#2 of 15	Official leaderboard
InfographicVQA	89.5%	#4 of 26	Lab model card
RefAdv-S	67.1%	#3 of 19	Official leaderboard
MuirBench	80.1%	#2 of 12	Lab model card
MM-MT-Bench	8.50	#1 of 6	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	92.4%	#5 of 29	Lab model card
MathVista	85.8%	#13 of 70	Lab model card
OCRBench v2	66.8%	#5 of 22	Lab model card
RefSpatial-Bench	69.9%	#4 of 16	Lab model card
MathVerse	85%	#3 of 12	Lab model card
EmbSpatialBench	84.3%	#5 of 16	Lab model card
WorldVQA	22.3%	#7 of 22	Official leaderboard
VideoMME v2 (with subtitles)	28.1%	#10 of 31	Official leaderboard
MMStar	78.7%	#12 of 36	Lab model card
RoboSpatial-Home	73.9%	#1 of 3	Lab model card
BLINK	67.1%	#10 of 29	Lab model card
MVBench	75.2%	#8 of 23	Lab model card
LogicVista	72.2%	#5 of 14	Lab model card
CiteVQA	22.5%	#6 of 16	Official leaderboard
OCRBench	87.5%	#13 of 34	Lab model card
MMBench-EN v1.1	90.6%	#9 of 23	Lab model card
AI2D	89.2%	#21 of 52	Lab model card
MMLongBench-Doc	56.2%	#7 of 17	Official leaderboard
RealWorldQA	81.3%	#16 of 38	Lab model card
CC-OCR	81.5%	#6 of 14	Lab model card
VSI-Bench	60%	#3 of 7	Lab model card
DynaMath	82.8%	#10 of 23	Lab model card
ZeroBench	27.7%	#14 of 31	Lab model card
CountBench	93.7%	#14 of 28	Lab model card
Charades-STA	63.5%	#2 of 4	Lab model card
MIA-Bench	92.7%	#1 of 2	Lab model card
MATH-Vision (MathVision)	74.6%	#26 of 51	Lab model card
MMVU	71.1%	#14 of 26	Lab model card
ODinW-13	43.2%	#7 of 13	Lab model card
We-Math	74.8%	#7 of 13	Lab model card
MLVU	83.8%	#12 of 22	Lab model card
MADQA	60.3%	#10 of 18	Official leaderboard
SUNRGBD	34.9	#5 of 9	Lab model card
CharXiv (DQ)	90.5%	#4 of 7	Lab model card
VisuLogic	34.4%	#4 of 7	Lab model card
MMMU-Pro	69.3%	#52 of 90	Lab model card
Video Reality Test	61.9%	#3 of 5	Official leaderboard
Video-MME	79%	#31 of 51	Lab model card
Vision Arena	1,188	#86 of 136	Official leaderboard
Video-MMMU	80%	#23 of 36	Lab model card
VisualPuzzles	57.2%	#2 of 3	Lab model card
BabyVision	22.2%	#19 of 27	Official leaderboard
V*	85.9%	#12 of 17	Lab model card
ERQA	52.5%	#26 of 36	Lab model card
LVBench	63.6%	#23 of 31	Lab model card
GeoBench	59%	#37 of 48	Official leaderboard
DISBench	8.20%	#7 of 9	Official leaderboard
VLMs Are Blind	79.5%	#11 of 14	Lab model card
CharXiv	66.1%	#59 of 75	Lab model card
VisFactor	20.4%	#22 of 26	Official leaderboard
Hypersim	11.0	#9 of 9	Lab model card
ARKitScenes	53.7%	#1 of 1	Lab model card
MathVision-Wildphoto	63.8%	#1 of 1	Lab model card

Other

Benchmark	Score		Rank	Source
LiveBench	79.6%		#2 of 27	Lab model card

Its lab's numbers put it at 140.6, number 173 of 404

The Self-Reported Index fits one score to the 60 capability results Qwen 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 139.6 to 141.7. How we compute it.

ZebraLogic 97.3
DocVQA 96.5
ScreenSpot 95.4
CountBench 93.7
MMLU-Redux 93.7
RefCOCO / RefCOCO+ / RefCOCOg 92.4
MMBench-EN v1.1 90.6
MMLU 90.6
ChartQA 90.3
AIME 2025 89.7
InfographicVQA 89.5
AI2D 89.2
IFEval 88.2
OCRBench 87.5
V* 85.9
MathVista 85.8
MathVerse 85
EmbSpatialBench 84.3
MLVU 83.8
MMLU-Pro 83.8
DynaMath 82.8
CC-OCR 81.5
RealWorldQA 81.3
MMMU 80.6
MMLU-ProX 80.6
MuirBench 80.1
Video-MMMU 80
INCLUDE 80
VLMs Are Blind 79.5
Multi-IF 79.1
Video-MME 79
MMStar 78.7
HMMT 2025 77.4
GPQA Diamond 77.1
MVBench 75.2
We-Math 74.8
Arena-Hard v2.0 74.8
MATH-Vision (MathVision) 74.6
LogicVista 72.2
Berkeley Function Calling Leaderboard (BFCL) 71.9
MMVU 71.1
LiveCodeBench v6 70.1
MMMU-Pro 69.3
BLINK 67.1
OCRBench v2 66.8
HallusionBench 66.7
CharXiv 66.1
SuperGPQA 64.3
LVBench 63.6
AndroidWorld 62
ScreenSpot-Pro 61.8
SimpleVQA 61.3
PolyMath 57.8
MMLongBench-Doc 56.2
SimpleQA 44.4
ODinW-13 43.2
OSWorld 38.1
ZeroBench 27.7
OJBench 27.5
Humanity's Last Exam 13.6

Its lab's numbers next to 7 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
MMMU	80.6% MMMU_VAL	78.7% Leaderboard	+1.9 pts
MMMU-Pro	69.3% MMMU_Pro	68.1% Leaderboard	+1.2 pts
MATH-Vision (MathVision)	74.6% MathVision	74.6% Leaderboard, thinking	Same
Math-VR	66.8% Math-VR (Table 2)	66.8% Leaderboard, thinking	Same
MMLongBench-Doc	56.2% MMLongBench-Doc	56.2% Leaderboard, VL; 22B activated (235B total)	Same
RoboSpatial-Home	73.9% RoboSpatialHome	73.9% Leaderboard, thinking	Same
VisuLogic	34.4% VisuLogic	34.4% Leaderboard, thinking	Same

Its launch post

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

Qwen3-VL launch post (qwen.ai blog) Sep 23, 2025

MMMU_VAL 80.6
MMMU_Pro 69.3
MathVista_mini 85.8
MathVision 74.6
MathVerse_mini 85
Best of ZEROBench 4.0 / ZEROBench_Sub 27.7 27.7
VisuLogic 34.4
MMBench_EN_V1.1_dev 90.6
RealWorldQA 81.3
MMStar 78.7
SimpleVQA 61.3
HallusionBench 66.7
MM_MT_Bench 8.5
MIABench 92.7
MMLongBench-Doc 56.2
DocVQA_TEST 96.5
InfoVQA_TEST 89.5
AI2D_TEST 89.2
OCRBench 875.0 (out of 1000) 87.5
OCRBenchV2 (en/ch) 66.8 / 63.5 66.8
CC_OCR 81.5
CharXiv(RQ) 66.1
RefCOCO-avg 92.4
CountBench 93.7
ODinW13 43.2
ARKitScenes 53.7
Hypersim 11
SUNRGBD 34.9
Objectron 71.2
BLINK 67.1
MUIRBENCH 80.1
ERQA 52.5
EmbSpatialBench 84.3
RefSpatialBench 69.9
RoboSpatialHome 73.9
VideoMME(w/o sub) 79
MLVU 83.8
LVBench 63.6
CharadesSTA 63.5
VideoMMMU 80
ScreenSpot 95.4
ScreenSpot Pro 61.8
OSWorldG 68.3
OSWorld 38.1
Design2Code 93.4
MMLU 90.6
MMLU-Pro 83.8
MMLU-Redux 93.7
SuperGPQA 64.3
SimpleQA 44.4
AIME25 89.7
HMMT25 77.4
LCB v6 70.1
ZebraLogic 97.3
LiveBench1125 79.6
HLE 13.6
SIFO (SIFO-multiturn 71.1) 77.3
IFEval 88.2
Creative Writing v3 85.7
WritingBench 86.7
BFCL 71.9
MultiIF 79.1
MMLU-ProX 80.6
INCLUDE 80

Its system card

Every result in the longer documents the lab published with it. 25 of them aren't in the launch post.

Qwen3-VL Technical Report Nov 26, 2025

Capabilities

MMMU 80.6
MMMU-Pro 69.3
MathVista 85.8
MATH-Vision (MathVision) 74.6
MathVision-Wildphoto 63.8
We-Math 74.8
MathVerse 85
DynaMath 82.8
Math-VR 66.8
ZeroBench 4
VLMs Are Blind 79.5
LogicVista 72.2
VisuLogic 34.4
VisualPuzzles 57.2
MMBench-EN v1.1 88.8
MMBench-CN v1.1 88.6
RealWorldQA 81.3
MMStar 78.7
SimpleVQA 61.3
HallusionBench 66.7
MM-MT-Bench 8.5
MIA-Bench 92.7
DocVQA 96.5
InfographicVQA 89.5
AI2D 89.2
ChartQA 90.3
OCRBench 87.5
OCRBench v2
CC-OCR 81.5
OmniDocBench
CharXiv (DQ) 90.5
CharXiv 66.1
MMLongBench-Doc 56.2
RefCOCO / RefCOCO+ / RefCOCOg 92.1
CountBench 93.7
ODinW-13 43.2
ARKitScenes 53.7
Hypersim 11
SUNRGBD 34.9
ERQA 52.5
VSI-Bench 60
EmbSpatialBench 84.3
RefSpatial-Bench 69.9
RoboSpatial-Home 73.9
BLINK 67.1
MuirBench 80.1
MVBench 75.2
Video-MME 79
MLVU 83.8
LVBench 63.6
Charades-STA 63.5
Video-MMMU 80
MMVU 71.1
V* 85.9
HR-Bench
Design2Code 93.4
ChartMimic 78.4
ScreenSpot-Pro 61.8
OSWorld-G 68.3
AndroidWorld 62
OSWorld 38.1
Windows Agent Arena 32.1
MMLU-Pro 83.8
MMLU-Redux 93.7
GPQA Diamond 77.1
SuperGPQA 64.3
AIME 2025 89.7
HMMT 2025 77.4
LiveBench 79.6
LiveCodeBench v6 70.1
Codeforces rating 1964
OJBench 27.5
IFEval 88.2
Arena-Hard v2.0 74.8
Creative Writing v3 85.7
WritingBench 86.7
Berkeley Function Calling Leaderboard (BFCL) 71.8
tau2-bench
Multi-IF 79.1
MMLU-ProX 80.6
INCLUDE 80
PolyMath 57.8