Which is better, Qwen3.5 397B A17B or Qwen3.6 27B?

Across 53 shared benchmarks tracked on Vector Wire, Qwen3.5 397B A17B leads on 32 and Qwen3.6 27B leads on 21 — Qwen3.5 397B A17B 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 397B A17B vs Qwen3.6 27B

AlibabavsAlibaba53 shared benchmarks32–21 head-to-head

Benchmark	Qwen3.5 397B A17B	Qwen3.6 27B
AA Agentic Index	53.3	27.5
AA Intelligence	34.3	37.7
AA-LCR	72.7	73.3
AA-Omniscience	-30.8	-20
aime_2026	94.2	94.1
Artificial Analysis Coding Index	48.2	53.7
C-Eval	93	91.4
CC-OCR	82	81.2
charxiv_rq	80.8	78.4
Claw Eval (pass@3)	48.1	60.6
Claw-Eval Avg	70.7	72.4
critpt	1.7	1.1
DynaMath	86.3	85.6
ERQA	67.5	62.5
gdpval	35.8	32
GPQA Diamond	89.3	87.8
HLE	29	24
HMMT 2025	92.7	90.7
hmmt_feb_2025	94.8	93.8
hmmt_feb_2026	87.9	84.3
hmmt_nov_2025	92.7	90.7
IFBench	78.8	69.1
imo_answer_bench	80.9	80.8
LiveCodeBench v6	83.6	83.9
MLVU	86.7	86.6
mmlu_redux	94.9	93.5
MMLU-Pro	87.8	86.2
MMMU	85	82.9
MMMU-Pro	79	75.8
MMStar	83.8	81.4
MV-Bench	77.6	75.5
nl2repo	32.2	36.2
OmniScience Accuracy	30.8	19.6
OmniScience Non-Hallucination	17.3	50.7
QwenClawBench	51.8	53.4
QwenWebBench	1186	1487
RealWorldQA	83.9	84.1
scicode	42	39.8
SimpleVQA	67.1	56.1
SkillsBench Avg5	30	48.2
supergpqa	70.4	66
SWE-bench Multilingual	69.3	71.3
SWE-bench Pro	50.9	53.5
SWE-bench Verified	76.4	77.2
TauBench V3 - Banking	13.4	16.7
Terminal-Bench 2.0	52.5	59.3
Terminal-Bench 2.1	51.3	60.7
Terminal-Bench Hard	40.9	34.8
V*	95.8	94.7
VideoMME (w sub.)	87.5	87.7
VideoMMMU	84.7	84.4
τ²-Bench Telecom (AA run)	95.6	94.2
τ³-Bench	13.4	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.