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

Across 87 shared benchmarks tracked on Vector Wire, Qwen3.5 27B leads on 65 and Qwen3.5 35B A3B leads on 20 — 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 Qwen3.5 35B A3B

AlibabavsAlibaba87 shared benchmarks65–20 head-to-head

Benchmark	Qwen3.5 27B	Qwen3.5 35B A3B
AA Agentic Index	54.6	44.1
AA Intelligence	34.6	29.9
AA-LCR	72.3	68.3
AA-Omniscience	-44	-48.1
AI2D	92.9	92.6
aime_2026	92.6	93.3
Artificial Analysis Coding Index	34.9	37
baby_vision	44.6	38.4
bfcl_v4	68.5	67.3
browsecomp	61	61
browsecomp_zh	62.1	69.5
C-Eval	90.5	90.2
CC-OCR	81	80.7
charxiv_rq	79.5	77.5
Claw Eval (pass@3)	46.2	51
Claw-Eval Avg	64.3	65.4
coding_arena_elo	1357	1250
critpt	0.9	0.9
DeepPlanning	22.6	22.8
EmbSpatial-Bench	84.5	0.8
ERQA	60.5	64.8
gdpval	33	20.3
GPQA Diamond	85.8	84.5
HallusionBench	70	67.9
HLE	48.5	47.4
HMMT 2025	89.8	89.2
hmmt_feb_2025	92	89
hmmt_feb_2026	84.3	81.8
hmmt_nov_2025	89.8	89.2
IFBench	76.5	72.5
ifeval	95	91.9
imo_answer_bench	79.9	76.8
include	81.6	79.7
itbenchSre	35.5	21.5
LiveCodeBench v6	80.7	74.6
longbench_v2	60.6	59
LVBench	73.6	71.4
mathvision	86	83.9
MathVista	87.8	86.2
MCP Atlas	68.4	62.4
mcpmark	36.3	27
MLVU	85.9	85.6
MMBench EN-DEV-v1.1	92.6	91.5
mmlu_prox	82.2	81
mmlu_redux	93.2	93.3
MMLU-Pro	86.1	85.3
mmmlu	85.9	85.2
MMMU	82.3	81.4
MMMU-Pro	75	75.1
MMStar	81	81.9
MMVU	73.3	72.3
multichallenge	60.8	60
MV-Bench	74.6	74.8
nl2repo	27.3	20.5
OCRBench	89.4	91
OJBench	40.1	36
OmniDocBench 1.5	88.9	89.3
OmniScience Accuracy	20.7	20.1
OmniScience Non-Hallucination	24.9	14.6
OSWorld-Verified	56.2	54.5
QwenClawBench	52.2	47.7
QwenWebBench	1068	978
RealWorldQA	83.7	84.1
RefSpatial-Bench	67.7	0.6
scicode	39.5	37.7
screenspot_pro_no_tools	70.3	68.6
Seal-0	47.2	41.4
SimpleVQA	56	58.3
SkillsBench Avg5	27.2	4.4
supergpqa	65.6	63.4
SWE-bench Multilingual	69.3	60.3
SWE-bench Pro	51.2	44.6
SWE-bench Verified	75	70
Terminal-Bench 2.0	41.6	40.5
Terminal-Bench Hard	32.6	26.5
tool_decathlon	31.5	28.7
vectara_answer_rate	99.8	99.8
vectara_avg_summary_length	94.4	94.9
vectara_factual_consistency	87.9	89.5
vectara_hallucination_rate ↓	12.1	10.5
VideoMMMU	82.3	80.4
VITA-Bench	41.9	31.9
WideSearch	66.4	59.1
ZeroBench	0.1	0.1
ZEROBench_sub	36.2	34.1
τ²-Bench Telecom (AA run)	93.9	89.2
τ³-Bench	68.4	68.9

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.