Qwen3.5-2B

Qwen3.5-2B.

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

Qwen3.5-2B

Open weights

123.7 #325 of 404

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

MMMU-Pro 50.3% #79 /90
IFBench 41.3% #72 /80
AA-LCR (Artificial Analysis Long Context Reasoning) 25.6% #37 /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	66.5%	#132 of 167	Lab model card
Global PIQA	58.9%	#35 of 44	Official leaderboard
IFEval	78.6%	#108 of 135	Lab model card
HallusionBench	58%	#25 of 31	Lab model card
AgentIF (2026-10 setting)	55.8%	#20 of 24	Official leaderboard
C-Eval	73.2%	#39 of 46	Lab model card
MMMLU	63.1%	#62 of 71	Lab model card
IFMTBench	43.1%	#7 of 8	Official leaderboard
IFBench	41.3%	#72 of 80	Lab model card
IFStruct	33.1%	#9 of 10	Official leaderboard
INCLUDE	55.4%	#28 of 31	Lab model card
MAXIFE	60.6%	#10 of 11	Lab model card
NOVA-63	46.4%	#10 of 11	Lab model card
MMLU-Redux	79.6%	#41 of 45	Lab model card
MMLU-ProX	52.3%	#21 of 23	Lab model card
KINA	20.5%	#34 of 37	Official leaderboard
SimpleVQA	38.5%	#28 of 30	Lab model card
LongBench v2	38.7%	#33 of 35	Lab model card
SuperGPQA	37.5%	#33 of 35	Lab model card
MultiChallenge	33.7%	#40 of 42	Lab model card
AA-LCR (Artificial Analysis Long Context Reasoning)	25.6%	#37 of 38	Lab model card
QIMMA	33.8%	#25 of 25	Official leaderboard
BOUQuET	19.2	#18 of 18	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
PHYSICS-Verified	19.4%	#15 of 19	Official leaderboard
GPQA Diamond	51.6%	#239 of 302	Lab model card
PolyMath	26.1%	#22 of 24	Lab model card
HMMT 2025	22.9%	#54 of 54	Lab model card
HMMT Nov 2025	19.6%	#28 of 28	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
ScreenSpot-Pro	54.5%	#26 of 32	Lab model card
tau2-bench	48.8%	#73 of 79	Lab model card
BFCL V4	43.6%	#28 of 30	Lab model card

Professional work

Benchmark	Score	Rank	Source
MedXpertQA-MM	26.9%	#18 of 21	Lab model card
PMC-VQA	48.8%	#9 of 10	Lab model card
SLAKE	74.4%	#9 of 10	Lab model card
MMEnterprise-Bench	25%	#9 of 9	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
MathVista	76.7%	#33 of 70	Lab model card
DynaMath	73.6%	#12 of 23	Lab model card
CountBench	91.4%	#17 of 28	Lab model card
RealWorldQA	74.5%	#24 of 38	Lab model card
MMStar	71.7%	#25 of 36	Lab model card
Video-MME	75.6%	#37 of 51	Lab model card
MMLongBench-Doc	45.4%	#14 of 19	Lab model card
ZeroBench	17.1%	#23 of 31	Lab model card
MMMU	64.2%	#78 of 103	Lab model card
OCRBench	84.5%	#26 of 34	Lab model card
MLVU	76.2%	#17 of 22	Lab model card
Hypersim	12.4	#7 of 9	Lab model card
MMBench-EN v1.1	83.3%	#18 of 23	Lab model card
GeoGuesser environment (FineEnvs)	0.28	#11 of 14	Official leaderboard
AI2D	83.3%	#42 of 52	Lab model card
ExtractBench	63.4%	#22 of 27	Official leaderboard
LVBench	57.1%	#26 of 31	Lab model card
RefCOCO / RefCOCO+ / RefCOCOg	84.8%	#25 of 29	Lab model card
CharXiv	58.8%	#65 of 75	Lab model card
Nuscene	6.90	#7 of 8	Lab model card
MMMU-Pro	50.3%	#79 of 90	Lab model card
SUNRGBD	28.7	#8 of 9	Lab model card
MVBench	64.9%	#21 of 23	Lab model card
Video-MMMU	62.1%	#33 of 36	Lab model card
ODinW-13	35.9%	#12 of 13	Lab model card
CC-OCR	72.9%	#13 of 14	Lab model card
VLMs Are Blind	75.8%	#13 of 14	Lab model card
EmbSpatialBench	77.9%	#15 of 16	Lab model card
RefSpatial-Bench	32.9%	#15 of 16	Lab model card
ERQA	43.8%	#34 of 36	Lab model card
MMVU	48.6%	#25 of 26	Lab model card
ParseBench	27.3%	#44 of 44	Official leaderboard

Safety and security

Benchmark	Score		Rank	Source
nocot-bench (NCRI)	35.5		#180 of 185	Official leaderboard

Its lab's numbers put it at 123.7, number 325 of 404

The Self-Reported Index fits one score to the 51 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 121.9 to 125.5. How we compute it.

CountBench 91.4
RefCOCO / RefCOCO+ / RefCOCOg 84.8
OCRBench 84.5
MMBench-EN v1.1 83.3
AI2D 83.3
MMLU-Redux 79.6
IFEval 78.6
EmbSpatialBench 77.9
MathVista 76.7
MLVU 76.2
VLMs Are Blind 75.8
Video-MME 75.6
RealWorldQA 74.5
SLAKE 74.4
DynaMath 73.6
C-Eval 73.2
CC-OCR 72.9
MMStar 71.7
Global PIQA 69.3
MMLU-Pro 66.5
MVBench 64.9
MMMU 64.2
MMMLU 63.1
Video-MMMU 62.1
MAXIFE 60.6
CharXiv 58.8
HallusionBench 58
LVBench 57.1
INCLUDE 55.4
ScreenSpot-Pro 54.5
MMLU-ProX 52.3
GPQA Diamond 51.6
MMMU-Pro 50.3
tau2-bench 48.8
PMC-VQA 48.8
MMVU 48.6
NOVA-63 46.4
MMLongBench-Doc 45.4
BFCL V4 43.6
IFBench 41.3
LongBench v2 38.7
SimpleVQA 38.5
SuperGPQA 37.5
ODinW-13 35.9
MultiChallenge 33.7
MedXpertQA-MM 26.9
PolyMath 26.1
AA-LCR (Artificial Analysis Long Context Reasoning) 25.6
HMMT 2025 22.9
HMMT Nov 2025 19.6
ZeroBench 17.1

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
MMLU-Pro	66.5% MMLU-Pro (Thinking)	55.3% Leaderboard	+11 pts
Global PIQA	69.3% Global PIQA	58.9% Leaderboard, exact_match,strict_match; instruct	+10 pts

Its launch post

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

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

MMLU-Pro (Thinking) 66.5
MMLU-Redux (Thinking) 79.6
C-Eval (Thinking) 73.2
SuperGPQA (Thinking) 37.5
IFEval (Thinking) 78.6
MMMLU (Thinking) 63.1
GPQA 51.6
IFBench 41.3
MultiChallenge 33.7
AA-LCR 25.6
LongBench v2 38.7
HMMT Feb 25 22.9
HMMT Nov 25 19.6
BFCL-V4 43.6
TAU2-Bench 48.8
MMLU-ProX 52.3
NOVA-63 46.4
INCLUDE 55.4
Global PIQA 69.3
PolyMATH 26.1
WMT24++ 45.8
MAXIFE 60.6
MMMU 64.2/64.2 (with CI / without CI) 64.2
MMMU-Pro 50.3/47.7 (with CI / without CI) 50.3
Mathvista(mini) 76.7/73.9 (with CI / without CI) 76.7
DynaMath 73.6/69.6 (with CI / without CI) 73.6
Best of ZEROBench / ZEROBench_sub 17.1/18.6 17.1
VlmsAreBlind 75.8/74.3 (with CI / without CI) 75.8
RealWorldQA 74.5/71.2 (with CI / without CI) 74.5
MMStar 71.7/68.0 (with CI / without CI) 71.7
MMBench EN-DEV-v1.1 83.3/81.3 (with CI / without CI) 83.3
SimpleVQA 38.5/39.5 (with CI / without CI) 38.5
HallusionBench 58.0/51.3 (with CI / without CI) 58
MMLongBench-Doc 45.4/38.8 (with CI / without CI) 45.4
AI2D_TEST 83.3/81.5 (with CI / without CI) 83.3
CC-OCR 72.9/75.8 (with CI / without CI) 72.9
OmniDocBench1.5 79.8/80.9 (with CI / without CI) 79.8
CharXiv(RQ) 58.8/52.6 (with CI / without CI) 58.8
OCRBench 84.5/85.4 (with CI / without CI) 84.5
RefCOCO(avg) 84.8/84.3 (with CI / without CI) 84.8
CountBench 91.4/86.8 (with CI / without CI) 91.4
ODInW13 35.9/40.5 (with CI / without CI) 35.9
ERQA 43.8/33.0 (with CI / without CI) 43.8
EmbSpatialBench 77.9/66.4 (with CI / without CI) 77.9
RefSpatialBench 32.9/30.0 (with CI / without CI) 32.9
Hypersim 12.4/12.4 (with CI / without CI) 12.4
SUNRGBD 28.7/25.6 (with CI / without CI) 28.7
Nuscene 6.9/8.5 (with CI / without CI) 6.9
Best of VideoMME(w sub.) / VideoMME(w/o sub.) 75.6/-- 75.6
VideoMMMU 62.1/-- (with CI / without CI) 62.1
MLVU 76.2/-- (with CI / without CI) 76.2
MVBench 64.9/-- (with CI / without CI) 64.9
LVBench 57.1/-- (with CI / without CI) 57.1
MMVU 48.6/-- (with CI / without CI) 48.6
SLAKE 74.4/67.5 (with CI / without CI) 74.4
PMC-VQA 48.8/54.0 (with CI / without CI) 48.8
MedXpertQA-MM 26.9/19.1 (with CI / without CI) 26.9
ScreenSpot Pro --/54.5 54.5