Which is better, DeepSeek-V3.2 or Qwen3.6 35B A3B?

Across 37 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 25 and Qwen3.6 35B A3B leads on 12 — DeepSeek-V3.2 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V3.2 vs Qwen3.6 35B A3B

DeepSeekvsAlibaba37 shared benchmarks25–12 head-to-head

Benchmark	DeepSeek-V3.2	Qwen3.6 35B A3B
AA Agentic Index	39.8	21.6
AA Intelligence	32.8	32.1
AA-LCR	70.7	66.7
AA-Omniscience	-22.5	-22.2
aime_2026	95.1	92.7
Artificial Analysis Coding Index	44.2	41.9
browsecomp	51.4	48.7
critpt	2.9	0.3
gdpval	18.8	27.8
GPQA (unspecified)	79.9	86
GPQA Diamond	84	86
HLE	40.8	22.2
HMMT 2025	90.2	89.1
hmmt_feb_2025	92.5	90.7
hmmt_feb_2026	84.1	83.6
hmmt_nov_2025	90.2	89.1
IFBench	61	64.4
imo_answer_bench	78.3	78.9
LiveCodeBench v6	83.3	80.4
MCP Atlas	62.2	62.8
mcpmark	38	37
mmlu_redux	93.7	93.3
MMLU-Pro	86	85.6
OmniScience Accuracy	33	18.8
OmniScience Non-Hallucination	17.3	49.5
scicode	39	35.8
SWE-bench Multilingual	70.2	67.2
SWE-bench Pro	15.6	49.5
SWE-bench Verified	73.1	73.4
TauBench V3 - Banking	18.8	9.3
Terminal-Bench 2.0	46.4	51.5
Terminal-Bench 2.1	46.8	44.9
Terminal-Bench Hard	35.6	34.8
tool_decathlon	35.2	26.9
toolathlon	35.2	26.9
τ²-Bench Telecom (AA run)	90.6	95.3
τ³-Bench	69.2	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.