Qwen3.5-0.8B

Qwen3.5-0.8B.

Qwen Released February 24th, 2026 Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Qwen Feb 2026

Qwen3.5-0.8B

Open weights

111.8 #372 of 404

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

MMMU-Pro 31.2% #87 /90
IFBench 21% #78 /80
AA-LCR (Artificial Analysis Long Context Reasoning) 4.70% #38 /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
HallusionBench	53.1%	#28 of 31	Lab model card
Global PIQA	52.5%	#41 of 44	Official leaderboard
MMLU-ProX	34.6%	#22 of 23	Lab model card
QIMMA	33.9%	#24 of 25	Official leaderboard
MMLU-Pro	42.3%	#162 of 167	Lab model card
KINA	16.7%	#36 of 37	Official leaderboard
IFBench	21%	#78 of 80	Lab model card
MultiChallenge	18.9%	#41 of 42	Lab model card
MMMLU	44.3%	#70 of 71	Lab model card
IFEval	44%	#135 of 135	Lab model card
C-Eval	50.5%	#46 of 46	Lab model card
MMLU-Redux	59.5%	#45 of 45	Lab model card
AA-LCR (Artificial Analysis Long Context Reasoning)	4.70%	#38 of 38	Lab model card
LongBench v2	26.1%	#35 of 35	Lab model card
SuperGPQA	21.3%	#35 of 35	Lab model card
INCLUDE	40.6%	#31 of 31	Lab model card
SimpleVQA	31.3%	#30 of 30	Lab model card
CulturalRiddles	0.40%	#20 of 20	Official leaderboard
MAXIFE	39.2%	#11 of 11	Lab model card
NOVA-63	42.4%	#11 of 11	Lab model card
IFStruct	15.5%	#10 of 10	Official leaderboard
IFMTBench	28%	#8 of 8	Official leaderboard

Math and science

Benchmark	Score		Rank	Source
GPQA Diamond	11.9%		#302 of 302	Lab model card
PolyMath	8.20%		#24 of 24	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
ScreenSpot-Pro	46.5%	#29 of 32	Lab model card
tau2-bench	11.6%	#79 of 79	Lab model card
BFCL V4	25.3%	#30 of 30	Lab model card

Professional work

Benchmark	Score	Rank	Source
MedXpertQA-MM	17.1%	#21 of 21	Lab model card
PMC-VQA	40.4%	#10 of 10	Lab model card
SLAKE	62.6%	#10 of 10	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
ZeroBench	12.9%	#24 of 31	Lab model card
MathVista	62.2%	#60 of 70	Lab model card
CountBench	77%	#24 of 28	Lab model card
DynaMath	49.9%	#20 of 23	Lab model card
ExtractBench	46.6%	#24 of 27	Official leaderboard
Hypersim	11.9	#8 of 9	Lab model card
RealWorldQA	63.4%	#34 of 38	Lab model card
MMMU	49%	#96 of 103	Lab model card
Video-MME	63.8%	#48 of 51	Lab model card
MMStar	58.3%	#34 of 36	Lab model card
MMLongBench-Doc	33.6%	#18 of 19	Lab model card
MMBench-EN v1.1	69.9%	#22 of 23	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	79.3%	#28 of 29	Lab model card
MMMU-Pro	31.2%	#87 of 90	Lab model card
OCRBench	74.5%	#33 of 34	Lab model card
ParseBench	28.4%	#43 of 44	Official leaderboard
CharXiv	41.3%	#74 of 75	Lab model card
AI2D	69.9%	#52 of 52	Lab model card
ERQA	34.5%	#36 of 36	Lab model card
Video-MMMU	44.3%	#36 of 36	Lab model card
LVBench	45.1%	#31 of 31	Lab model card
MMVU	34.3%	#26 of 26	Lab model card
MVBench	55.8%	#23 of 23	Lab model card
MLVU	65.6%	#22 of 22	Lab model card
EmbSpatialBench	68.6%	#16 of 16	Lab model card
RefSpatial-Bench	23.5%	#16 of 16	Lab model card
CC-OCR	63.2%	#14 of 14	Lab model card
GeoGuesser environment (FineEnvs)	0.16	#14 of 14	Official leaderboard
VLMs Are Blind	59.4%	#14 of 14	Lab model card
ODinW-13	31.6%	#13 of 13	Lab model card
SUNRGBD	26.1	#9 of 9	Lab model card
Nuscene	5.70	#8 of 8	Lab model card

Safety and security

Benchmark	Score		Rank	Source
nocot-bench (NCRI)	24.9		#183 of 185	Official leaderboard

Its lab's numbers put it at 111.8, number 372 of 404

The Self-Reported Index fits one score to the 49 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 109.6 to 114.0. How we compute it.

RefCOCO / RefCOCO+ / RefCOCOg 79.3
CountBench 77
OCRBench 74.5
MMBench-EN v1.1 69.9
AI2D 69.9
EmbSpatialBench 68.6
MLVU 65.6
Video-MME 63.8
RealWorldQA 63.4
CC-OCR 63.2
SLAKE 62.6
MathVista 62.2
MMLU-Redux 59.5
Global PIQA 59.4
VLMs Are Blind 59.4
MMStar 58.3
MVBench 55.8
HallusionBench 53.1
C-Eval 50.5
DynaMath 49.9
MMMU 49
ScreenSpot-Pro 46.5
LVBench 45.1
MMMLU 44.3
Video-MMMU 44.3
IFEval 44
NOVA-63 42.4
MMLU-Pro 42.3
CharXiv 41.3
INCLUDE 40.6
PMC-VQA 40.4
MAXIFE 39.2
MMLU-ProX 34.6
MMVU 34.3
MMLongBench-Doc 33.6
ODinW-13 31.6
SimpleVQA 31.3
MMMU-Pro 31.2
LongBench v2 26.1
BFCL V4 25.3
SuperGPQA 21.3
IFBench 21
MultiChallenge 18.9
MedXpertQA-MM 17.1
ZeroBench 12.9
GPQA Diamond 11.9
tau2-bench 11.6
PolyMath 8.2
AA-LCR (Artificial Analysis Long Context Reasoning) 4.7

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
MMLU-Pro	42.3% MMLU-Pro (Thinking)	29.7% Leaderboard	+13 pts
Global PIQA	59.4% Global PIQA	52.5% Leaderboard, exact_match,strict_match; instruct	+6.9 pts

Its launch post

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

Qwen3.5-0.8B model card (Hugging Face), Qwen3.5 small series Mar 2, 2026

MMLU-Pro (Thinking) 42.3
MMLU-Redux (Thinking) 59.5
C-Eval (Thinking) 50.5
SuperGPQA (Thinking) 21.3
IFEval (Thinking) 44
MMMLU (Thinking) 44.3
GPQA 11.9
IFBench 21
MultiChallenge 18.9
AA-LCR 4.7
LongBench v2 26.1
BFCL-V4 25.3
TAU2-Bench 11.6
MMLU-ProX 34.6
NOVA-63 42.4
INCLUDE 40.6
Global PIQA 59.4
PolyMATH 8.2
WMT24++ 27.2
MAXIFE 39.2
MMMU 49/47.4 (with CI / without CI) 49
MMMU-Pro 31.2/31.4 (with CI / without CI) 31.2
Mathvista(mini) 62.2/58.6 (with CI / without CI) 62.2
DynaMath 49.9/46.5 (with CI / without CI) 49.9
Best of ZEROBench / ZEROBench_sub 12.9/11.4 12.9
VlmsAreBlind 59.4/57.3 (with CI / without CI) 59.4
RealWorldQA 63.4/61.6 (with CI / without CI) 63.4
MMStar 58.3/55.9 (with CI / without CI) 58.3
MMBench EN-DEV-v1.1 69.9/68.0 (with CI / without CI) 69.9
SimpleVQA 31.3/30.4 (with CI / without CI) 31.3
HallusionBench 53.1/46.7 (with CI / without CI) 53.1
MMLongBench-Doc 33.6/28.1 (with CI / without CI) 33.6
AI2D_TEST 69.9/68.7 (with CI / without CI) 69.9
CC-OCR 63.2/66.7 (with CI / without CI) 63.2
OmniDocBench1.5 61.0/70.6 (with CI / without CI) 61
CharXiv(RQ) 41.3/38.2 (with CI / without CI) 41.3
OCRBench 74.5/79.1 (with CI / without CI) 74.5
RefCOCO(avg) 79.3/77.8 (with CI / without CI) 79.3
CountBench 77.0/68.6 (with CI / without CI) 77
ODInW13 31.6/33.2 (with CI / without CI) 31.6
ERQA 34.5/23.8 (with CI / without CI) 34.5
EmbSpatialBench 68.6/54.6 (with CI / without CI) 68.6
RefSpatialBench 23.5/21.7 (with CI / without CI) 23.5
Hypersim 11.9/11.0 (with CI / without CI) 11.9
SUNRGBD 26.1/23.3 (with CI / without CI) 26.1
Nuscene 5.7/7.0 (with CI / without CI) 5.7
Best of VideoMME(w sub.) / VideoMME(w/o sub.) 63.8/-- 63.8
VideoMMMU 44.3/-- (with CI / without CI) 44.3
MLVU 65.6/-- (with CI / without CI) 65.6
MVBench 55.8/-- (with CI / without CI) 55.8
LVBench 45.1/-- (with CI / without CI) 45.1
MMVU 34.3/-- (with CI / without CI) 34.3
SLAKE 62.6/59.5 (with CI / without CI) 62.6
PMC-VQA 40.4/45.5 (with CI / without CI) 40.4
MedXpertQA-MM 17.1/25.3 (with CI / without CI) 17.1
ScreenSpot Pro --/46.5 46.5