Qwen3.5-122B-A10B

Qwen3.5-122B-A10B places in the top three on 10 benchmarks.

Qwen Released February 24th, 2026 Compare with others

103

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Qwen Feb 2026

Qwen3.5-122B-A10B

Open weights

146.7 #119 of 404

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

MMMU-Pro 76.9% #31 /90
IFBench 76.1% #20 /80
AA-LCR (Artificial Analysis Long Context Reasoning) 66.9% #14 /38

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-Pro	86.7%	#13 of 167	Lab model card
IFEval	93.4%	#14 of 135	Lab model card
QIMMA	66.3%	#3 of 25	Official leaderboard
MMLU-Redux	94%	#7 of 45	Lab model card
C-Eval	91.9%	#8 of 46	Lab model card
SEA-HELM	74.7	#15 of 71	Official leaderboard
IFBench	76.1%	#20 of 80	Lab model card
SuperGPQA	67.1%	#9 of 35	Lab model card
MMLU-ProX	82.2%	#6 of 23	Lab model card
HallusionBench	67.6%	#9 of 31	Lab model card
INCLUDE	82.8%	#9 of 31	Lab model card
MMMLU	86.7%	#21 of 71	Lab model card
KINA	38.9%	#11 of 37	Official leaderboard
LMArena (Elo)	1,416	#109 of 335	Official leaderboard
Humanity's Last Exam	47.5%	#51 of 154	Lab model card
Judgemark v2.1	81.0	#14 of 42	Official leaderboard
NOVA-63	58.6%	#4 of 11	Lab model card
AA-LCR (Artificial Analysis Long Context Reasoning)	66.9%	#14 of 38	Lab model card
MultiChallenge	61.5%	#17 of 42	Lab model card
LongBench v2	60.2%	#15 of 35	Lab model card
MAXIFE	87.9%	#6 of 11	Lab model card
Global PIQA	88.4%	#8 of 14	Lab model card
SimpleVQA	61.7%	#18 of 30	Lab model card
Longform Creative Writing (EQ-Bench)	54.3	#78 of 119	Official leaderboard
Extended NYT Connections	51.7%	#71 of 103	Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset)	11.2%	#73 of 104	Official leaderboard
Seal-0	44.1%	#6 of 7	Lab model card
WideSearch	60.5%	#17 of 19	Lab model card

Math and science

Benchmark	Score	Rank	Source
GPQA Diamond	86.6%	#63 of 302	Lab model card
PolyMath	68.9%	#7 of 24	Lab model card
HMMT 2025	91.4%	#21 of 54	Lab model card
DTBench	84.3%	#73 of 180	Official leaderboard
PHYSICS-Verified	80.7%	#8 of 19	Official leaderboard
HMMT Nov 2025	90.3%	#14 of 28	Lab model card

Coding

Benchmark	Score	Rank	Source
Codeforces rating	2,100	#18 of 47	Lab model card
LiveCodeBench v6	78.9%	#35 of 90	Lab model card
SWE-bench Verified	72%	#67 of 155	Lab model card
OJBench	39.5%	#8 of 16	Lab model card
FullStackBench	60.7%	#4 of 8	Lab model card
Code Arena	1,360	#97 of 128	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
SnakeBench	33	#22 of 188	Official leaderboard
BFCL V4	72.2%	#5 of 30	Lab model card
GameWorld	45.2%	#5 of 22	Official leaderboard
BrowseComp-ZH	69.9%	#7 of 19	Lab model card
ScreenSpot-Pro	70.4%	#13 of 32	Lab model card
tau2-bench	79.5%	#40 of 79	Lab model card
LisanBench	392.7	#73 of 133	Official leaderboard
VITA-Bench	33.6%	#9 of 16	Lab model card
AndroidWorld	66.4%	#13 of 23	Lab model card
DeepPlanning	24.1%	#6 of 10	Lab model card
Terminal-Bench 2.0	49.4%	#39 of 63	Lab model card
BrowseComp	63.8%	#54 of 81	Lab model card
AutoResearchBench	3.88%	#9 of 13	Official leaderboard
Mystery Game Puzzles	17%	#60 of 84	Epoch AI run
OSWorld-Verified	58%	#32 of 41	Lab model card
MobileWorld	29.3%	#15 of 18	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
MedXpertQA-MM	67.3%	#6 of 21	Lab model card
PMC-VQA	63.3%	#3 of 10	Lab model card
SLAKE	81.6%	#3 of 10	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
MMMU	83.9%	#7 of 103	Lab model card
MMBench-EN v1.1	92.8%	#2 of 23	Lab model card
MathVista	87.4%	#7 of 70	Lab model card
RefAdv-S	67.2%	#2 of 19	Official leaderboard
OCRBench	92.1%	#4 of 34	Lab model card
GSM8K-V	42.1%	#3 of 24	Official leaderboard
MLVU	87.3%	#3 of 22	Lab model card
MMStar	82.9%	#5 of 36	Lab model card
DynaMath	85.9%	#4 of 23	Lab model card
MVBench	76.6%	#4 of 23	Lab model card
ZeroBench	36.2%	#6 of 31	Lab model card
GeoGuesser environment (FineEnvs)	0.44	#3 of 14	Official leaderboard
AI2D	93.3%	#12 of 52	Lab model card
CountBench	97%	#7 of 28	Lab model card
Nuscene	15.4	#2 of 8	Lab model card
Video-MME	87.3%	#13 of 51	Lab model card
RealWorldQA	85.1%	#10 of 38	Lab model card
MATH-Vision (MathVision)	86.2%	#15 of 51	Lab model card
MMVU	74.7%	#8 of 26	Lab model card
RefSpatial-Bench	69.3%	#5 of 16	Lab model card
SUNRGBD	36.2	#3 of 9	Lab model card
MMMU-Pro	76.9%	#31 of 90	Lab model card
CC-OCR	81.8%	#5 of 14	Lab model card
VLMs Are Blind	96.7%	#5 of 14	Lab model card
MMLongBench-Doc	59%	#7 of 19	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	91.3%	#11 of 29	Lab model card
VideoMME v2 (with subtitles)	26.7%	#12 of 31	Official leaderboard
V*	93.2%	#7 of 17	Lab model card
EmbSpatialBench	83.9%	#7 of 16	Lab model card
Hypersim	12.7	#4 of 9	Lab model card
LVBench	74.4%	#14 of 31	Lab model card
ODinW-13	44.5%	#6 of 13	Lab model card
Vision Arena	1,226	#63 of 136	Official leaderboard
ERQA	62%	#18 of 36	Lab model card
LingoQA	80.8%	#5 of 10	Lab model card
Video-MMMU	82%	#19 of 36	Lab model card
CharXiv	77.2%	#45 of 75	Lab model card
TIR-Bench	53.2%	#4 of 6	Lab model card
BabyVision	40.2%	#19 of 26	Lab model card
CiteVQA	14.8%	#12 of 16	Official leaderboard

Safety and security

Benchmark	Score		Rank	Source
nocot-bench (NCRI)	69.4		#103 of 185	Official leaderboard

Other

Benchmark	Score	Rank	Source
Conceptual Reasoning Index	45.7	#85 of 151	Official leaderboard
LMCA	32.2%	#86 of 147	Official leaderboard
ACCoRD	58.0	#96 of 152	Official leaderboard

Its lab's numbers put it at 146.7, number 119 of 404

The Self-Reported Index fits one score to the 68 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 145.7 to 147.6. How we compute it.

CountBench 97
VLMs Are Blind 96.7
MMLU-Redux 94
IFEval 93.4
AI2D 93.3
V* 93.2
MMBench-EN v1.1 92.8
OCRBench 92.1
C-Eval 91.9
HMMT 2025 91.4
RefCOCO / RefCOCO+ / RefCOCOg 91.3
HMMT Nov 2025 90.3
Global PIQA 88.4
MAXIFE 87.9
MathVista 87.4
Video-MME 87.3
MLVU 87.3
MMLU-Pro 86.7
MMMLU 86.7
GPQA Diamond 86.6
MATH-Vision (MathVision) 86.2
DynaMath 85.9
RealWorldQA 85.1
MMMU 83.9
EmbSpatialBench 83.9
MMStar 82.9
INCLUDE 82.8
MMLU-ProX 82.2
Video-MMMU 82
CC-OCR 81.8
SLAKE 81.6
LingoQA 80.8
tau2-bench 79.5
LiveCodeBench v6 78.9
CharXiv 77.2
MMMU-Pro 76.9
MVBench 76.6
IFBench 76.1
MMVU 74.7
LVBench 74.4
BFCL V4 72.2
SWE-bench Verified 72
ScreenSpot-Pro 70.4
BrowseComp-ZH 69.9
PolyMath 68.9
HallusionBench 67.6
MedXpertQA-MM 67.3
SuperGPQA 67.1
AA-LCR (Artificial Analysis Long Context Reasoning) 66.9
AndroidWorld 66.4
BrowseComp 63.8
PMC-VQA 63.3
SimpleVQA 61.7
MultiChallenge 61.5
FullStackBench 60.7
WideSearch 60.5
LongBench v2 60.2
MMLongBench-Doc 59
NOVA-63 58.6
OSWorld-Verified 58
Terminal-Bench 2.0 49.4
Humanity's Last Exam 47.5
ODinW-13 44.5
BabyVision 40.2
OJBench 39.5
ZeroBench 36.2
VITA-Bench 33.6
DeepPlanning 24.1

Its lab's numbers next to 2 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
MATH-Vision (MathVision)	86.2% MathVision	86.2% Leaderboard	Same
MMLU-Pro	86.7% MMLU-Pro	86.7% Leaderboard	Same

Its launch post

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

Qwen3.5-122B-A10B model card (Hugging Face), Qwen3.5 medium series Feb 24, 2026

MMLU-Pro 86.7
MMLU-Redux 94
C-Eval 91.9
SuperGPQA 67.1
IFEval 93.4
IFBench 76.1
MultiChallenge 61.5
AA-LCR 66.9
LongBench v2 60.2
Best of HLE w/ CoT 25.3 / HLE w/ tool 47.5 47.5
GPQA Diamond 86.6
HMMT Feb 25 91.4
HMMT Nov 25 90.3
SWE-bench Verified 72
Terminal Bench 2 49.4
LiveCodeBench v6 78.9
CodeForces (Qwen query set) 2100
OJBench 39.5
Mean of FullStackBench en 62.6 / FullStackBench zh 58.7 60.7
BFCL-V4 72.2
TAU2-Bench (airline fixes from Claude Opus 4.5 system card) 79.5
VITA-Bench 33.6
DeepPlanning 24.1
Browsecomp (context folding) 63.8
Browsecomp-zh 69.9
WideSearch 60.5
Seal-0 44.1
MMMLU 86.7
MMLU-ProX (29 languages) 82.2
NOVA-63 58.6
INCLUDE 82.8
Global PIQA 88.4
PolyMATH 68.9
WMT24++ (XCOMET-XXL, 55 languages) 78.3
MAXIFE 87.9
MMMU 83.9
MMMU-Pro 76.9
MathVision 86.2
Mathvista(mini) 87.4
DynaMath 85.9
Best of ZEROBench 9 / ZEROBench_sub 36.2 36.2
VlmsAreBlind 96.7
BabyVision 40.2 / 34.5 (with CI / without CI) 40.2
RealWorldQA 85.1
MMStar 82.9
MMBenchEN-DEV-v1.1 92.8
SimpleVQA 61.7
HallusionBench 67.6
OmniDocBench1.5 89.8
CharXiv(RQ) 77.2
MMLongBench-Doc 59
CC-OCR 81.8
AI2D_TEST 93.3
OCRBench 92.1
ERQA 62
CountBench 97
RefCOCO(avg) 91.3
ODInW13 44.5
EmbSpatialBench 83.9
RefSpatialBench 69.3
LingoQA 80.8
Hypersim 12.7
SUNRGBD 36.2
Nuscene 15.4
Best of VideoMME(w sub.) 87.3 / VideoMME(w/o sub.) 83.9 87.3
VideoMMMU 82
MLVU 87.3
MVBench 76.6
LVBench 74.4
MMVU 74.7
ScreenSpot Pro 70.4
OSWorld-Verified 58
AndroidWorld 66.4
TIR-Bench 53.2 / 42.5 (with CI / without CI) 53.2
V* 93.2 / 90.1 (with CI / without CI) 93.2
SLAKE 81.6
PMC-VQA 63.3
MedXpertQA-MM 67.3