Which is better, DeepSeek-V4-Pro or Qwen3.5 35B A3B?

Across 45 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 42 and Qwen3.5 35B A3B leads on 3 — DeepSeek-V4-Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V4-Pro vs Qwen3.5 35B A3B

DeepSeekvsAlibaba45 shared benchmarks42–3 head-to-head

Benchmark	DeepSeek-V4-Pro	Qwen3.5 35B A3B
AA Agentic Index	63.3	44.1
AA Intelligence	53	29.9
AA-LCR	70	68.3
AA-Omniscience	-10.6	-48.1
AgentWorldBench - Android	55.2	53.2
AgentWorldBench - MCP	63.3	57.9
AgentWorldBench - OS	63.7	56.3
AgentWorldBench - Overall	53	47.7
AgentWorldBench - Search	27.6	26
AgentWorldBench - SWE	59.4	47.6
AgentWorldBench - Terminal	51.3	46.1
AgentWorldBench - Web	50.3	47.1
aime_2026	96.7	93.3
Artificial Analysis Coding Index	59.4	37
browsecomp	83.4	61
coding_arena_elo	1582	1250
critpt	13	0.9
gdpval	49	20.3
GPQA Diamond	90.5	84.5
HLE	48.2	47.4
hmmt_feb_2026	95.2	81.8
hmmt_nov_2025	94.4	89.2
IFBench	76.5	72.5
imo_answer_bench	89.8	76.8
itbenchSre	38.3	21.5
MCP Atlas	74.2	62.4
MMLU-Pro	87.5	85.3
nl2repo	38.5	20.5
OmniScience Accuracy	43	20.1
OmniScience Non-Hallucination	12.2	14.6
scicode	50	37.7
SWE-bench Multilingual	76.2	60.3
SWE-bench Pro	55.4	44.6
SWE-bench Verified	80.6	70
TauBench V3 - Banking	30.1	4.9
Terminal-Bench 2.0	67.9	40.5
Terminal-Bench 2.1	72.1	40.8
Terminal-Bench Hard	46.2	26.5
tool_decathlon	52.8	28.7
vectara_answer_rate	97.2	99.8
vectara_avg_summary_length	153.8	94.9
vectara_factual_consistency	91.4	89.5
vectara_hallucination_rate ↓	8.6	10.5
τ²-Bench Telecom (AA run)	96.2	89.2
τ³-Bench	25.8	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.