Which is better, DeepSeek-V4-Flash or DeepSeek-V4-Pro?

Across 64 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Flash leads on 26 and DeepSeek-V4-Pro leads on 38 — 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-Flash vs DeepSeek-V4-Pro

DeepSeekvsDeepSeek64 shared benchmarks26–38 head-to-head

Benchmark	DeepSeek-V4-Flash	DeepSeek-V4-Pro
AA Agentic Index	61.3	63.3
AA Intelligence	51.8	53
AA-LCR	74.3	70
AA-Omniscience	-14.3	-10.6
Agents' Last Exam	25.2	16.5
aime_2026	95.8	96.7
Apex (Pass@1)	33	38.3
Apex Shortlist (Pass@1)	85.7	90.2
arena_elo	1439	1457
arena_text_factuality	1434	1450
Artificial Analysis Coding Index	69.1	59.4
AutomationBench (Public)	25.1	12.8
AutomationBench Public	25.1	12.8
browsecomp	73.2	83.4
Chinese SimpleQA (C-SimpleQA)	78.9	77.7
Codeforces	3052	3206
coding_arena_elo	1431	1582
CorpusQA 1M (ACC)	60.5	62
critpt	16.6	13
cybergym	76.7	52.7
DeepSWE	54.4	12.8
DSBench-FullStack	68.7	41.8
DSBench-Hard	59.6	31.1
gdpval	52.9	49
GDPval-AA (Elo)	1395	1554
GDPval-AA v2	1189	1307
Global-MMLU-Lite	88.4	89.3
GPQA Diamond	90.8	90.5
HLE	38.6	48.2
HLE (with tools)	45.1	48.2
HLE (wo / w tools)	51.5	48.2
hmmt_feb_2026	94.8	95.2
IFBench	79.2	76.5
imo_answer_bench	88.4	89.8
itbenchSre	31.5	38.3
livebench_agentic_coding	46.8	42.6
livebench_coding	75	70
livebench_data_analysis	79.3	74.5
livebench_instruction_following	65.5	62.4
livebench_language	79.2	78.1
livebench_math	86.8	90.7
livebench_reasoning	86.6	82.7
LiveCodeBench	91.6	93.5
MCP Atlas	69	74.2
MMLU-Pro	86.4	87.5
MRCR 1M (MMR)	78.7	83.5
nl2repo	54.2	38.5
OmniScience Accuracy	40.4	43
OmniScience Non-Hallucination	15.6	12.2
scicode	49.9	50
simplebench	46.3	50.9
simpleqa_verified	45	57.9
strongreject	97.4	98.6
SWE-bench Multilingual	73.3	76.2
SWE-bench Pro	52.6	55.4
SWE-bench Verified	79	80.6
Tau 3 Banking	22.9	26
TauBench V3 - Banking	39.4	30.1
Terminal-Bench 2.0	56.9	67.9
Terminal-Bench 2.1	82.7	72.1
Terminal-Bench Hard	38.6	46.2
toolathlon	70.3	51.8
τ²-Bench Telecom (AA run)	95.6	96.2
τ³-Bench	22.9	25.8

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.