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

Across 62 shared benchmarks tracked on Vector Wire, Qwen3.5 27B leads on 36 and Qwen3.6 35B A3B leads on 25 — Qwen3.5 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 35B A3B

AlibabavsAlibaba62 shared benchmarks36–25 head-to-head

Benchmark	Qwen3.5 27B	Qwen3.6 35B A3B
AA Agentic Index	54.6	21.6
AA Intelligence	34.6	32.1
AA-LCR	72.3	66.7
AA-Omniscience	-44	-22.2
AI2D	92.9	92.7
aime_2026	92.6	92.7
Artificial Analysis Coding Index	34.9	41.9
browsecomp	61	48.7
C-Eval	90.5	90
CC-OCR	81	81.9
charxiv_rq	79.5	78
Claw Eval (pass@3)	46.2	50
Claw-Eval Avg	64.3	68.7
critpt	0.9	0.3
DeepPlanning	22.6	25.9
EmbSpatial-Bench	84.5	0.8
gdpval	33	27.8
GPQA Diamond	85.8	86
HallusionBench	70	69.8
HLE	48.5	22.2
HMMT 2025	89.8	89.1
hmmt_feb_2025	92	90.7
hmmt_feb_2026	84.3	83.6
hmmt_nov_2025	89.8	89.1
IFBench	76.5	64.4
imo_answer_bench	79.9	78.9
LiveCodeBench v6	80.7	80.4
LVBench	73.6	71.4
MathVista	87.8	86.4
MCP Atlas	68.4	62.8
mcpmark	36.3	37
MLVU	85.9	86.2
MMBench EN-DEV-v1.1	92.6	92.8
mmlu_redux	93.2	93.3
MMLU-Pro	86.1	85.6
MMMU	82.3	81.7
MMMU-Pro	75	75.3
MV-Bench	74.6	74.6
nl2repo	27.3	29.4
OmniDocBench 1.5	88.9	89.9
OmniScience Accuracy	20.7	18.8
OmniScience Non-Hallucination	24.9	49.5
QwenClawBench	52.2	52.6
QwenWebBench	1068	1397
RealWorldQA	83.7	85.3
RefSpatial-Bench	67.7	0.6
scicode	39.5	35.8
SimpleVQA	56	58.9
SkillsBench Avg5	27.2	28.7
supergpqa	65.6	64.7
SWE-bench Multilingual	69.3	67.2
SWE-bench Pro	51.2	49.5
SWE-bench Verified	75	73.4
Terminal-Bench 2.0	41.6	51.5
Terminal-Bench Hard	32.6	34.8
tool_decathlon	31.5	26.9
VideoMMMU	82.3	83.7
VITA-Bench	41.9	35.6
WideSearch	66.4	60.1
ZEROBench_sub	36.2	34.4
τ²-Bench Telecom (AA run)	93.9	95.3
τ³-Bench	68.4	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.