Which is better, Qwen3.5 27B or Qwen3 VL 4B (Reasoning)?

Across 39 shared benchmarks tracked on Vector Wire, Qwen3.5 27B leads on 39 and Qwen3 VL 4B (Reasoning) leads on 0 — 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 VL 4B (Reasoning)

AlibabavsAlibaba39 shared benchmarks39–0 head-to-head

Benchmark	Qwen3.5 27B	Qwen3 VL 4B (Reasoning)
AA Agentic Index	54.6	14.4
AA Intelligence	34.6	7.7
AA-LCR	72.3	23
AA-Omniscience	-44	-68.4
AI2D	92.9	84.9
Artificial Analysis Coding Index	34.9	6.7
critpt	0.9	0
EmbSpatial-Bench	84.5	80.7
ERQA	60.5	47.3
gdpval	33	13.8
GPQA Diamond	85.8	64.1
HallusionBench	70	64.1
HLE	48.5	4.6
HMMT 2025	89.8	53.1
IFBench	76.5	36.6
ifeval	95	82.6
include	81.6	64.6
LiveCodeBench v6	80.7	51.3
LVBench	73.6	53.5
mathvision	86	60
MathVista	87.8	79.5
mmlu_prox	82.2	65
mmlu_redux	93.2	86
MMLU-Pro	86.1	73.6
MMMU-Pro	75	57
MMStar	81	73.2
MV-Bench	74.6	69.3
OCRBench	89.4	81.6
OmniScience Accuracy	20.7	12.2
OmniScience Non-Hallucination	24.9	8.2
OSWorld-Verified	56.2	31.4
RealWorldQA	83.7	73.2
RefSpatial-Bench	67.7	45.3
scicode	39.5	17.1
screenspot_pro_no_tools	70.3	49.2
supergpqa	65.6	46.8
Terminal-Bench Hard	32.6	1.5
VideoMMMU	82.3	69.4
τ²-Bench Telecom (AA run)	93.9	15.5

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.