Which is better, Qwen3.5 35B A3B or Qwen3.6 27B?

Across 61 shared benchmarks tracked on Vector Wire, Qwen3.5 35B A3B leads on 10 and Qwen3.6 27B leads on 50 — Qwen3.6 27B leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Qwen3.5 35B A3B vs Qwen3.6 27B

AlibabavsAlibaba61 shared benchmarks10–50 head-to-head

Benchmark	Qwen3.5 35B A3B	Qwen3.6 27B
AA Agentic Index	44.1	27.5
AA Intelligence	29.9	37.7
AA-LCR	68.3	73.3
AA-Omniscience	-48.1	-20
aime_2026	93.3	94.1
Artificial Analysis Coding Index	37	53.7
baby_vision	38.4	28.9
C-Eval	90.2	91.4
CC-OCR	80.7	81.2
charxiv_rq	77.5	78.4
Claw Eval (pass@3)	51	60.6
Claw-Eval Avg	65.4	72.4
critpt	0.9	1.1
EmbSpatial-Bench	0.8	84.6
ERQA	64.8	62.5
gdpval	20.3	32
GPQA Diamond	84.5	87.8
hellaswag	85.6	90
HLE	47.4	24
HMMT 2025	89.2	90.7
hmmt_feb_2025	89	93.8
hmmt_feb_2026	81.8	84.3
hmmt_nov_2025	89.2	90.7
humaneval	66.5	78.2
IFBench	72.5	69.1
imo_answer_bench	76.8	80.8
LiveCodeBench v6	74.6	83.9
mathvision	83.9	85.1
MathVista	86.2	87.4
MLVU	85.6	86.6
MMBench EN-DEV-v1.1	91.5	92.3
mmlu_redux	93.3	93.5
MMLU-Pro	85.3	86.2
MMMU	81.4	82.9
MMMU-Pro	75.1	75.8
MMStar	81.9	81.4
MV-Bench	74.8	75.5
nl2repo	20.5	36.2
OCRBench	91	89.4
OmniDocBench 1.5	89.3	89.4
OmniScience Accuracy	20.1	19.6
OmniScience Non-Hallucination	14.6	50.7
OSWorld-Verified	54.5	63.9
QwenClawBench	47.7	53.4
QwenWebBench	978	1487
RealWorldQA	84.1	84.1
RefSpatial-Bench	0.6	70
scicode	37.7	39.8
SimpleVQA	58.3	56.1
SkillsBench Avg5	4.4	48.2
supergpqa	63.4	66
SWE-bench Multilingual	60.3	71.3
SWE-bench Pro	44.6	53.5
SWE-bench Verified	70	77.2
TauBench V3 - Banking	4.9	16.7
Terminal-Bench 2.0	40.5	59.3
Terminal-Bench 2.1	40.8	60.7
Terminal-Bench Hard	26.5	34.8
VideoMMMU	80.4	84.4
τ²-Bench Telecom (AA run)	89.2	94.2
τ³-Bench	68.9	15.3

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.