Which is better, DeepSeek-V4-Flash or GPT-5.2?

Across 46 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Flash leads on 23 and GPT-5.2 leads on 23 — DeepSeek-V4-Flash 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 GPT-5.2

DeepSeekvsOpenAI46 shared benchmarks23–23 head-to-head

Benchmark	DeepSeek-V4-Flash	GPT-5.2
AA Agentic Index	61.3	60.2
AA Intelligence	51.8	43.3
AA-LCR	74.3	79.3
AA-Omniscience	-14.3	0.3
aime_2026	95.8	98.3
arc_agi_1	84	86.2
ARC-AGI-2	46	52.9
arena_elo	1439	1437
arena_text_factuality	1434	1442
Artificial Analysis Coding Index	69.1	48.7
browsecomp	73.2	65.8
BrowseComp (context management)	73.2	70
coding_arena_elo	1431	1418
critpt	16.6	11.6
gdpval	52.9	48.3
GDPval-AA (Elo)	1395	1462
GPQA Diamond	90.8	92.4
HLE	38.6	45.5
HLE (with tools)	45.1	45.5
hmmt_feb_2026	94.8	97
IFBench	79.2	75.4
imo_answer_bench	88.4	86.3
livebench_agentic_coding	46.8	50.3
livebench_coding	75	76.1
livebench_data_analysis	79.3	78.2
livebench_instruction_following	65.5	61.8
livebench_language	79.2	79.8
livebench_math	86.8	93.2
livebench_reasoning	86.6	83.2
LiveCodeBench	91.6	89
MCP Atlas	69	68
MMLU-Pro	86.4	87
OmniScience Accuracy	40.4	44.3
OmniScience Non-Hallucination	15.6	38.4
scicode	49.9	52.1
simplebench	46.3	45.8
simpleqa_verified	45	38.9
SWE-bench Multilingual	73.3	72
SWE-bench Pro	52.6	55.6
SWE-bench Verified	79	80
TauBench V3 - Banking	39.4	11.1
Terminal-Bench 2.0	56.9	54
Terminal-Bench Hard	38.6	62.2
toolathlon	70.3	46.3
τ²-Bench Telecom (AA run)	95.6	98.7
τ³-Bench	22.9	12

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.