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

Across 61 shared benchmarks tracked on Vector Wire, Qwen3.5 27B leads on 12 and Qwen3.6 27B leads on 47 — 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 27B vs Qwen3.6 27B

AlibabavsAlibaba61 shared benchmarks12–47 head-to-head

Benchmark	Qwen3.5 27B	Qwen3.6 27B
AA Agentic Index	54.6	27.5
AA Intelligence	34.6	37.7
AA-LCR	72.3	73.3
AA-Omniscience	-44	-20
aime_2026	92.6	94.1
Artificial Analysis Coding Index	34.9	53.7
baby_vision	44.6	28.9
C-Eval	90.5	91.4
CC-OCR	81	81.2
charxiv_rq	79.5	78.4
Claw Eval (pass@3)	46.2	60.6
Claw-Eval Avg	64.3	72.4
critpt	0.9	1.1
DynaMath	87.7	85.6
EmbSpatial-Bench	84.5	84.6
ERQA	60.5	62.5
gdpval	33	32
GPQA Diamond	85.8	87.8
HLE	48.5	24
HMMT 2025	89.8	90.7
hmmt_feb_2025	92	93.8
hmmt_feb_2026	84.3	84.3
hmmt_nov_2025	89.8	90.7
IFBench	76.5	69.1
imo_answer_bench	79.9	80.8
LiveCodeBench v6	80.7	83.9
mathvision	86	85.1
MathVista	87.8	87.4
MLVU	85.9	86.6
MMBench EN-DEV-v1.1	92.6	92.3
mmlu_redux	93.2	93.5
MMLU-Pro	86.1	86.2
MMMU	82.3	82.9
MMMU-Pro	75	75.8
MMStar	81	81.4
MV-Bench	74.6	75.5
nl2repo	27.3	36.2
OCRBench	89.4	89.4
OmniDocBench 1.5	88.9	89.4
OmniScience Accuracy	20.7	19.6
OmniScience Non-Hallucination	24.9	50.7
OSWorld-Verified	56.2	63.9
QwenClawBench	52.2	53.4
QwenWebBench	1068	1487
RealWorldQA	83.7	84.1
RefSpatial-Bench	67.7	70
scicode	39.5	39.8
SimpleVQA	56	56.1
SkillsBench Avg5	27.2	48.2
supergpqa	65.6	66
SWE-bench Multilingual	69.3	71.3
SWE-bench Pro	51.2	53.5
SWE-bench Verified	75	77.2
Terminal-Bench 2.0	41.6	59.3
Terminal-Bench Hard	32.6	34.8
V*	93.7	94.7
VideoMME (w sub.)	87	87.7
VideoMMMU	82.3	84.4
VlmsAreBlind	96.9	97
τ²-Bench Telecom (AA run)	93.9	94.2
τ³-Bench	68.4	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.