Which is better, DeepSeek-V3.2 or Qwen3.5 35B A3B?

Across 44 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 32 and Qwen3.5 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 Qwen3.5 35B A3B

DeepSeekvsAlibaba44 shared benchmarks32–12 head-to-head

Benchmark	DeepSeek-V3.2	Qwen3.5 35B A3B
AA Agentic Index	39.8	44.1
AA Intelligence	32.8	29.9
AA-LCR	70.7	68.3
AA-Omniscience	-22.5	-48.1
aime_2026	95.1	93.3
Artificial Analysis Coding Index	44.2	37
browsecomp	51.4	61
browsecomp_zh	65	69.5
coding_arena_elo	1368	1250
critpt	2.9	0.9
gdpval	18.8	20.3
GPQA (unspecified)	79.9	84.2
GPQA Diamond	84	84.5
HLE	40.8	47.4
HMMT 2025	90.2	89.2
hmmt_feb_2025	92.5	89
hmmt_feb_2026	84.1	81.8
hmmt_nov_2025	90.2	89.2
IFBench	61	72.5
imo_answer_bench	78.3	76.8
LiveCodeBench v6	83.3	74.6
longbench_v2	59.8	59
MCP Atlas	62.2	62.4
mcpmark	38	27
mmlu_redux	93.7	93.3
MMLU-Pro	86	85.3
OmniScience Accuracy	33	20.1
OmniScience Non-Hallucination	17.3	14.6
scicode	39	37.7
Seal-0	49.5	41.4
SWE-bench Multilingual	70.2	60.3
SWE-bench Pro	15.6	44.6
SWE-bench Verified	73.1	70
TauBench V3 - Banking	18.8	4.9
Terminal-Bench 2.0	46.4	40.5
Terminal-Bench 2.1	46.8	40.8
Terminal-Bench Hard	35.6	26.5
tool_decathlon	35.2	28.7
vectara_answer_rate	92.6	99.8
vectara_avg_summary_length	62	94.9
vectara_factual_consistency	93.7	89.5
vectara_hallucination_rate ↓	6.3	10.5
τ²-Bench Telecom (AA run)	90.6	89.2
τ³-Bench	69.2	68.9

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.