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

Across 53 shared benchmarks tracked on Vector Wire, Qwen3.6 27B leads on 45 and Qwen3.6 35B A3B leads on 7 — 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 —

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

AlibabavsAlibaba53 shared benchmarks45–7 head-to-head

Benchmark	Qwen3.6 27B	Qwen3.6 35B A3B
AA Agentic Index	27.5	21.6
AA Intelligence	37.7	32.1
AA-LCR	73.3	66.7
AA-Omniscience	-20	-22.2
aime_2026	94.1	92.7
Artificial Analysis Coding Index	53.7	41.9
C-Eval	91.4	90
CC-OCR	81.2	81.9
charxiv_rq	78.4	78
Claw Eval (pass@3)	60.6	50
Claw-Eval Avg	72.4	68.7
critpt	1.1	0.3
EmbSpatial-Bench	84.6	0.8
gdpval	32	27.8
GPQA Diamond	87.8	86
HLE	24	22.2
HMMT 2025	90.7	89.1
hmmt_feb_2025	93.8	90.7
hmmt_feb_2026	84.3	83.6
hmmt_nov_2025	90.7	89.1
IFBench	69.1	64.4
imo_answer_bench	80.8	78.9
LiveCodeBench v6	83.9	80.4
MathVista	87.4	86.4
MLVU	86.6	86.2
MMBench EN-DEV-v1.1	92.3	92.8
mmlu_redux	93.5	93.3
MMLU-Pro	86.2	85.6
MMMU	82.9	81.7
MMMU-Pro	75.8	75.3
MV-Bench	75.5	74.6
nl2repo	36.2	29.4
OmniDocBench 1.5	89.4	89.9
OmniScience Accuracy	19.6	18.8
OmniScience Non-Hallucination	50.7	49.5
QwenClawBench	53.4	52.6
QwenWebBench	1487	1397
RealWorldQA	84.1	85.3
RefSpatial-Bench	70	0.6
scicode	39.8	35.8
SimpleVQA	56.1	58.9
SkillsBench Avg5	48.2	28.7
supergpqa	66	64.7
SWE-bench Multilingual	71.3	67.2
SWE-bench Pro	53.5	49.5
SWE-bench Verified	77.2	73.4
TauBench V3 - Banking	16.7	9.3
Terminal-Bench 2.0	59.3	51.5
Terminal-Bench 2.1	60.7	44.9
Terminal-Bench Hard	34.8	34.8
VideoMMMU	84.4	83.7
τ²-Bench Telecom (AA run)	94.2	95.3
τ³-Bench	15.3	67.2

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.