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

Across 51 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 30 and GPT-5.2 leads on 21 — 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 GPT-5.2

DeepSeekvsOpenAI51 shared benchmarks30–21 head-to-head

Benchmark	DeepSeek-V4-Pro	GPT-5.2
AA Agentic Index	63.3	60.2
AA Intelligence	53	43.3
AA-LCR	70	79.3
AA-Omniscience	-10.6	0.3
aime_2026	96.7	98.3
apexAgents	24.3	23
arena_elo	1457	1437
arena_text_factuality	1450	1442
Artificial Analysis Coding Index	59.4	48.7
browsecomp	83.4	65.8
coding_arena_elo	1582	1418
critpt	13	11.6
gdpval	49	48.3
GDPval-AA (Elo)	1554	1462
GPQA Diamond	90.5	92.4
HLE	48.2	45.5
HLE (with tools)	48.2	45.5
hmmt_feb_2026	95.2	97
hmmt_nov_2025	94.4	99.2
IFBench	76.5	75.4
imo_answer_bench	89.8	86.3
livebench	73.6	74.8
livebench_agentic_coding	42.6	50.3
livebench_coding	70	76.1
livebench_data_analysis	74.5	78.2
livebench_instruction_following	62.4	61.8
livebench_language	78.1	79.8
livebench_math	90.7	93.2
livebench_reasoning	82.7	83.2
LiveCodeBench	93.5	89
MCP Atlas	74.2	68
MMLU-Pro	87.5	87
OmniScience Accuracy	43	44.3
OmniScience Non-Hallucination	12.2	38.4
scicode	50	52.1
simplebench	50.9	45.8
simpleqa_verified	57.9	38.9
SWE-bench Multilingual	76.2	72
SWE-bench Pro	55.4	55.6
SWE-bench Verified	80.6	80
TauBench V3 - Banking	30.1	11.1
Terminal-Bench 2.0	67.9	54
Terminal-Bench Hard	46.2	62.2
tool_decathlon	52.8	46.3
toolathlon	51.8	46.3
vectara_answer_rate	97.2	100
vectara_avg_summary_length	153.8	186.3
vectara_factual_consistency	91.4	89.2
vectara_hallucination_rate ↓	8.6	10.8
τ²-Bench Telecom (AA run)	96.2	98.7
τ³-Bench	25.8	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.