Which is better, Qwen3.5 397B A17B or Qwen3.6 35B A3B?

Across 55 shared benchmarks tracked on Vector Wire, Qwen3.5 397B A17B leads on 47 and Qwen3.6 35B A3B leads on 8 — 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 35B A3B

AlibabavsAlibaba55 shared benchmarks47–8 head-to-head

Benchmark	Qwen3.5 397B A17B	Qwen3.6 35B A3B
AA Agentic Index	53.3	21.6
AA Intelligence	34.3	32.1
AA-LCR	72.7	66.7
AA-Omniscience	-30.8	-22.2
aime_2026	94.2	92.7
Artificial Analysis Coding Index	48.2	41.9
browsecomp	69	48.7
C-Eval	93	90
CC-OCR	82	81.9
charxiv_rq	80.8	78
Claw Eval (pass@3)	48.1	50
Claw-Eval Avg	70.7	68.7
critpt	1.7	0.3
DeepPlanning	34.3	25.9
gdpval	35.8	27.8
GDPval-AA v2	962	1015
GPQA Diamond	89.3	86
HLE	29	22.2
HMMT 2025	92.7	89.1
hmmt_feb_2025	94.8	90.7
hmmt_feb_2026	87.9	83.6
hmmt_nov_2025	92.7	89.1
IFBench	78.8	64.4
imo_answer_bench	80.9	78.9
LiveCodeBench v6	83.6	80.4
mcpmark	46.1	37
MLVU	86.7	86.2
mmlu_redux	94.9	93.3
MMLU-Pro	87.8	85.6
MMMU	85	81.7
MMMU-Pro	79	75.3
MV-Bench	77.6	74.6
nl2repo	32.2	29.4
OmniScience Accuracy	30.8	18.8
OmniScience Non-Hallucination	17.3	49.5
QwenClawBench	51.8	52.6
QwenWebBench	1186	1397
RealWorldQA	83.9	85.3
scicode	42	35.8
SimpleVQA	67.1	58.9
SkillsBench Avg5	30	28.7
supergpqa	70.4	64.7
SWE-bench Multilingual	69.3	67.2
SWE-bench Pro	50.9	49.5
SWE-bench Verified	76.4	73.4
TauBench V3 - Banking	13.4	9.3
Terminal-Bench 2.0	52.5	51.5
Terminal-Bench 2.1	51.3	44.9
Terminal-Bench Hard	40.9	34.8
toolathlon	38.3	26.9
VideoMMMU	84.7	83.7
VITA-Bench	49.7	35.6
WideSearch	74	60.1
τ²-Bench Telecom (AA run)	95.6	95.3
τ³-Bench	13.4	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.