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

Across 66 shared benchmarks tracked on Vector Wire, Qwen3.5 35B A3B leads on 16 and Qwen3.6 35B A3B leads on 48 — Qwen3.6 35B A3B 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 35B A3B vs Qwen3.6 35B A3B

AlibabavsAlibaba66 shared benchmarks16–48 head-to-head

Benchmark	Qwen3.5 35B A3B	Qwen3.6 35B A3B
AA Agentic Index	44.1	21.6
AA Intelligence	29.9	32.1
AA-LCR	68.3	66.7
AA-Omniscience	-48.1	-22.2
AI2D	92.6	92.7
aime_2026	93.3	92.7
Artificial Analysis Coding Index	37	41.9
browsecomp	61	48.7
C-Eval	90.2	90
CC-OCR	80.7	81.9
charxiv_rq	77.5	78
Claw Eval (pass@3)	51	50
Claw-Eval Avg	65.4	68.7
critpt	0.9	0.3
DeepPlanning	22.8	25.9
EmbSpatial-Bench	0.8	0.8
gdpval	20.3	27.8
GPQA (unspecified)	84.2	86
GPQA Diamond	84.5	86
HallusionBench	67.9	69.8
HLE	47.4	22.2
HMMT 2025	89.2	89.1
hmmt_feb_2025	89	90.7
hmmt_feb_2026	81.8	83.6
hmmt_nov_2025	89.2	89.1
IFBench	72.5	64.4
imo_answer_bench	76.8	78.9
LiveCodeBench v6	74.6	80.4
LVBench	71.4	71.4
MathVista	86.2	86.4
MCP Atlas	62.4	62.8
mcpmark	27	37
MLVU	85.6	86.2
MMBench EN-DEV-v1.1	91.5	92.8
mmlu_redux	93.3	93.3
MMLU-Pro	85.3	85.6
MMMU	81.4	81.7
MMMU-Pro	75.1	75.3
MV-Bench	74.8	74.6
nl2repo	20.5	29.4
ocrbench_v2	65.3	65.5
OmniDocBench 1.5	89.3	89.9
OmniScience Accuracy	20.1	18.8
OmniScience Non-Hallucination	14.6	49.5
QwenClawBench	47.7	52.6
QwenWebBench	978	1397
RealWorldQA	84.1	85.3
RefSpatial-Bench	0.6	0.6
scicode	37.7	35.8
SimpleVQA	58.3	58.9
SkillsBench Avg5	4.4	28.7
supergpqa	63.4	64.7
SWE-bench Multilingual	60.3	67.2
SWE-bench Pro	44.6	49.5
SWE-bench Verified	70	73.4
TauBench V3 - Banking	4.9	9.3
Terminal-Bench 2.0	40.5	51.5
Terminal-Bench 2.1	40.8	44.9
Terminal-Bench Hard	26.5	34.8
tool_decathlon	28.7	26.9
VideoMMMU	80.4	83.7
VITA-Bench	31.9	35.6
WideSearch	59.1	60.1
ZEROBench_sub	34.1	34.4
τ²-Bench Telecom (AA run)	89.2	95.3
τ³-Bench	68.9	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.