Which is better, DeepSeek-V3.2 or Qwen3.7 Max?

Across 38 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 4 and Qwen3.7 Max leads on 34 — Qwen3.7 Max 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 Qwen3.7 Max

DeepSeekvsAlibaba38 shared benchmarks4–34 head-to-head

Benchmark	DeepSeek-V3.2	Qwen3.7 Max
AA Agentic Index	39.8	30.9
AA Intelligence	32.8	46.7
AA-LCR	70.7	74.7
AA-Omniscience	-22.5	13.5
aime_2026	95.1	97
arena_text_factuality	1433	1479
Artificial Analysis Coding Index	44.2	66
coding_arena_elo	1368	1517
critpt	2.9	13.4
gdpval	18.8	38.5
GPQA Diamond	84	92.4
HealthBench	46.9	54.5
HLE	40.8	41.4
HLE (with tools)	40.8	53.5
hmmt_feb_2026	84.1	97.1
hmmt_nov_2025	90.2	95
IFBench	61	80.5
imo_answer_bench	78.3	90
LiveCodeBench v6	83.3	91.6
longbench_v2	59.8	65.3
MCP Atlas	62.2	76.4
mcpmark	38	60.8
mmlu_redux	93.7	95
MMLU-Pro	86	89.6
OmniScience Accuracy	33	31.1
OmniScience Non-Hallucination	17.3	74.4
PaperBench	47.1	64.8
scicode	39	53.5
simpleqa_verified	27.5	58.5
SWE-bench Multilingual	70.2	78.3
SWE-bench Pro	15.6	60.6
SWE-bench Verified	73.1	80.4
TauBench V3 - Banking	18.8	11.8
Terminal-Bench 2.0	46.4	69.7
Terminal-Bench 2.1	46.8	75
Terminal-Bench Hard	35.6	50.8
τ²-Bench Telecom (AA run)	90.6	94.7
τ³-Bench	69.2	10.9

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.