Which is better, DeepSeek-V4-Flash or Qwen3.7 Max?

Across 43 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Flash leads on 17 and Qwen3.7 Max leads on 26 — 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-V4-Flash vs Qwen3.7 Max

DeepSeekvsAlibaba43 shared benchmarks17–26 head-to-head

Benchmark	DeepSeek-V4-Flash	Qwen3.7 Max
AA Agentic Index	61.3	30.9
AA Intelligence	51.8	46.7
AA-LCR	74.3	74.7
AA-Omniscience	-14.3	13.5
aime_2026	95.8	97
arena_elo	1439	1474
arena_text_factuality	1434	1479
Artificial Analysis Coding Index	69.1	66
coding_arena_elo	1431	1517
critpt	16.6	13.4
DeepSWE	54.4	18
gdpval	52.9	38.5
GPQA Diamond	90.8	92.4
HLE	38.6	41.4
HLE (with tools)	45.1	53.5
hmmt_feb_2026	94.8	97.1
IFBench	79.2	80.5
imo_answer_bench	88.4	90
itbenchSre	31.5	42.5
livebench_agentic_coding	46.8	43.6
livebench_coding	75	74.2
livebench_data_analysis	79.3	71.8
livebench_instruction_following	65.5	74
livebench_language	79.2	79.7
livebench_math	86.8	85.3
livebench_reasoning	86.6	83.3
MCP Atlas	69	76.4
MMLU-Pro	86.4	89.6
nl2repo	54.2	47.2
OmniScience Accuracy	40.4	31.1
OmniScience Non-Hallucination	15.6	74.4
scicode	49.9	53.5
simplebench	46.3	70.4
simpleqa_verified	45	58.5
SWE-bench Multilingual	73.3	78.3
SWE-bench Pro	52.6	60.6
SWE-bench Verified	79	80.4
TauBench V3 - Banking	39.4	11.8
Terminal-Bench 2.0	56.9	69.7
Terminal-Bench 2.1	82.7	75
Terminal-Bench Hard	38.6	50.8
τ²-Bench Telecom (AA run)	95.6	94.7
τ³-Bench	22.9	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.