Which is better, DeepSeek-V4-Pro or Qwen3.5 397B A17B?

Across 50 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 38 and Qwen3.5 397B A17B leads on 12 — 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 397B A17B

DeepSeekvsAlibaba50 shared benchmarks38–12 head-to-head

Benchmark	DeepSeek-V4-Pro	Qwen3.5 397B A17B
AA Agentic Index	63.3	53.3
AA Intelligence	53	34.3
AA-LCR	70	72.7
AA-Omniscience	-10.6	-30.8
AgentWorldBench - Android	55.2	54.9
AgentWorldBench - MCP	63.3	68.3
AgentWorldBench - OS	63.7	60.9
AgentWorldBench - Overall	53	54.7
AgentWorldBench - Search	27.6	30.8
AgentWorldBench - SWE	59.4	64.4
AgentWorldBench - Terminal	51.3	55.3
AgentWorldBench - Web	50.3	48.5
aime_2026	96.7	94.2
apexAgents	24.3	15.3
arena_elo	1457	1442
arena_text_factuality	1450	1445
Artificial Analysis Coding Index	59.4	48.2
browsecomp	83.4	69
coding_arena_elo	1582	1399
critpt	13	1.7
gdpval	49	35.8
GDPval-AA v2	1307	962
Global-MMLU-Lite	89.3	90
GPQA Diamond	90.5	89.3
HLE	48.2	29
HLE (with tools)	48.2	48.3
hmmt_feb_2026	95.2	87.9
hmmt_nov_2025	94.4	92.7
IFBench	76.5	78.8
imo_answer_bench	89.8	80.9
itbenchSre	38.3	34.1
MMLU-Pro	87.5	87.8
nl2repo	38.5	32.2
OmniScience Accuracy	43	30.8
OmniScience Non-Hallucination	12.2	17.3
scicode	50	42
simpleqa_verified	57.9	26
strongreject	98.6	99.4
SWE-bench Multilingual	76.2	69.3
SWE-bench Pro	55.4	50.9
SWE-bench Verified	80.6	76.4
Tau 3 Banking	26	13.4
TauBench V3 - Banking	30.1	13.4
Terminal-Bench 2.0	67.9	52.5
Terminal-Bench 2.1	72.1	51.3
Terminal-Bench Hard	46.2	40.9
toolathlon	51.8	38.3
usamo_2026	60.7	36.3
τ²-Bench Telecom (AA run)	96.2	95.6
τ³-Bench	25.8	13.4

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.