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

Across 67 shared benchmarks tracked on Vector Wire, Qwen3.5 35B A3B leads on 6 and Qwen3.6 Plus leads on 61 — Qwen3.6 Plus 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 Plus

AlibabavsAlibaba67 shared benchmarks6–61 head-to-head

Benchmark	Qwen3.5 35B A3B	Qwen3.6 Plus
AA Agentic Index	44.1	29
AA Intelligence	29.9	40.5
AA-LCR	68.3	72.3
AA-Omniscience	-48.1	2.6
AgentWorldBench - Android	53.2	57.6
AgentWorldBench - MCP	57.9	55.3
AgentWorldBench - OS	56.3	60.3
AgentWorldBench - Overall	47.7	50.8
AgentWorldBench - Search	26	21.9
AgentWorldBench - SWE	47.6	59.1
AgentWorldBench - Terminal	46.1	50.6
AgentWorldBench - Web	47.1	50.8
AI2D	92.6	94.4
aime_2026	93.3	95.3
Artificial Analysis Coding Index	37	54.5
C-Eval	90.2	93.3
Claw Eval (pass@3)	51	58.7
coding_arena_elo	1250	1460
critpt	0.9	2.9
DeepPlanning	22.8	41.5
ERQA	64.8	65.7
gdpval	20.3	32
GPQA Diamond	84.5	90.4
HLE	47.4	28.8
HMMT 2025	89.2	94.6
hmmt_feb_2026	81.8	87.8
hmmt_nov_2025	89.2	94.6
IFBench	72.5	75.2
ifeval	91.9	94.3
imo_answer_bench	76.8	83.8
include	79.7	85.1
LiveCodeBench v6	74.6	87.1
longbench_v2	59	62
mathvision	83.9	88
MCP Atlas	62.4	74.1
mcpmark	27	48.2
MLVU	85.6	86.7
mmlu_prox	81	84.7
mmlu_redux	93.3	94.5
MMLU-Pro	85.3	88.5
mmmlu	85.2	89.5
MMMU	81.4	86
MMMU-Pro	75.1	78.8
MMStar	81.9	83.3
nl2repo	20.5	37.9
OmniDocBench 1.5	89.3	91.2
OmniScience Accuracy	20.1	26.4
OmniScience Non-Hallucination	14.6	68
OSWorld-Verified	54.5	62.5
RealWorldQA	84.1	85.4
scicode	37.7	40.7
screenspot_pro_no_tools	68.6	68.2
SimpleVQA	58.3	0.7
supergpqa	63.4	71.6
SWE-bench Multilingual	60.3	73.8
SWE-bench Pro	44.6	56.6
SWE-bench Verified	70	78.8
TauBench V3 - Banking	4.9	20.8
Terminal-Bench 2.0	40.5	61.6
Terminal-Bench 2.1	40.8	61.4
Terminal-Bench Hard	26.5	43.9
tool_decathlon	28.7	39.8
VideoMMMU	80.4	84
VITA-Bench	31.9	44.3
WideSearch	59.1	74.3
τ²-Bench Telecom (AA run)	89.2	97.7
τ³-Bench	68.9	70.7

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.