Keye-VL-2.0-30B-A3B

Keye-VL-2.0-30B-A3B ranks first on 2 of the 27 benchmarks with five or more models scored.

Kuaishou Released May 25th, 2026 Family Keye-VL The lab's model card Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Kuaishou May 2026

Keye-VL-2.0-30B-A3B

Open weights Mid-size

144.0 #142 of 404

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

BFCL V4 65.7% #15 /30
HallusionBench 73.5% #1 /31
VideoMME v2 (with subtitles) 42.4% #5 /6

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	73.5%	#1 of 31	Lab model card
IFEval	92.2%	#19 of 135	Lab model card
C-Eval	89.2%	#22 of 46	Lab model card
Humanity's Last Exam	22.9%	#97 of 154	Lab model card

Math and science

Benchmark	Score		Rank	Source
AIME 2025	86.7%		#66 of 143	Lab model card
AIME 2026	93.3%		#19 of 38	Lab model card

Coding

Benchmark	Score	Rank	Source
OJBench	71.5%	#1 of 16	Lab model card
LiveCodeBench v6	77.1%	#40 of 90	Lab model card
SWE-bench Verified	62%	#106 of 155	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
tau2-bench	82.6%	#28 of 79	Lab model card
BFCL V4	65.7%	#15 of 30	Lab model card
VITA-Bench	33.1%	#10 of 16	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
We-Math	75.5%	#4 of 13	Lab model card
MathVista	82.9%	#23 of 70	Lab model card
OmniDocBench v1.5	84.7	#3 of 9	Lab model card
MMMU	75.3%	#42 of 103	Lab model card
DynaMath	80.9%	#11 of 23	Lab model card
MMStar	75.9%	#18 of 36	Lab model card
PixMo-Count	83.5%	#1 of 2	Lab model card
LongVideoBench	74.1%	#6 of 11	Lab model card
EmbSpatialBench	83.2%	#9 of 16	Lab model card
AI2D	87.8%	#31 of 52	Lab model card
MLVU	82.8%	#14 of 22	Lab model card
Video-MME	78.3%	#33 of 51	Lab model card
OCRBench	85.7%	#22 of 34	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	88.1%	#19 of 29	Lab model card
Video-MMMU	80%	#24 of 36	Lab model card
VideoMME v2 (with subtitles)	42.4%	#5 of 6	Lab model card

Its lab's numbers put it at 144.0, number 142 of 404

The Self-Reported Index fits one score to the 25 capability results Kuaishou 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 141.2 to 146.8. How we compute it.

Its launch post

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

Keye-VL-2.0-30B-A3B model card (Hugging Face) May 25, 2026

LongVideoBench 74.1
Best of Video-MME-v2 ACC 64 frames 35.3 / 512 frames 42.4 42.4
Mean of ActivityNet-TimeLens 58.5 / QVHighlights-TimeLens 70.1 / Charades-TimeLens 58.4 62.3
MLVU 82.8
VideoMME (w/o sub.) 78.3
VideoMMMU 80
LiveCodeBench v6 77.1
OJBench 39.2
SWE-bench Verified 62
BFCL-V4 65.7
TAU2-Bench 82.6
VITA-Bench 33.1
WeMath 75.5
Mathvista (mini) 82.9
DynaMath 80.9
AIME2025 86.7
AIME2026 93.3
HLE w/ CoT 22.9
MMMU 75.3
MMStar 75.9
C-Eval 89.2
IFEval 92.2
HallusionBench 73.5
AI2D_TEST 87.8
OCRBench 85.7
OmniDocBench1.5 84.7
RefCOCO (avg) 88.1
Pixmo-Count 83.5
EmbSpatialBench 83.2

Its system card

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

Kwai Keye-VL-2.0 Technical Report Jun 9, 2026

Capabilities

Hallucination

HallusionBench 73.5