Which is better, DeepSeek-V4-Pro or GPT-5.4?

Across 66 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 15 and GPT-5.4 leads on 50 — GPT-5.4 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-5.4

DeepSeekvsOpenAI66 shared benchmarks15–50 head-to-head

Benchmark	DeepSeek-V4-Pro	GPT-5.4
AA Agentic Index	63.3	58.2
AA Intelligence	53	53.1
AA-LCR	70	77.7
AA-Omniscience	-10.6	5.8
AgentWorldBench - Android	55.2	60
AgentWorldBench - MCP	63.3	70.1
AgentWorldBench - OS	63.7	68.6
AgentWorldBench - Overall	53	58.3
AgentWorldBench - Search	27.6	37.3
AgentWorldBench - SWE	59.4	66.3
AgentWorldBench - Terminal	51.3	53.7
AgentWorldBench - Web	50.3	51.8
aime_2026	96.7	99.2
Apex (Pass@1)	38.3	54.1
Apex Shortlist (Pass@1)	90.2	78.1
apexAgents	24.3	33.3
arena_elo	1457	1477
arena_text_factuality	1450	1476
Artificial Analysis Coding Index	59.4	71.1
browsecomp	83.4	82.7
Chinese SimpleQA (C-SimpleQA)	77.7	76.8
Codeforces	3206	3168
coding_arena_elo	1582	1457
critpt	13	23.4
cybergym	52.7	66.3
gdpval	49	50.1
GDPval-AA (Elo)	1554	1674
GPQA Diamond	90.5	93
HLE	48.2	43.7
HLE (with tools)	48.2	52.1
hmmt_feb_2026	95.2	97.7
hmmt_nov_2025	94.4	95.8
IFBench	76.5	73.9
imo_answer_bench	89.8	91.4
itbenchSre	38.3	34.5
livebench	73.6	80.3
livebench_agentic_coding	42.6	53.8
livebench_coding	70	77.5
livebench_data_analysis	74.5	79.3
livebench_instruction_following	62.4	70.2
livebench_language	78.1	82.6
livebench_math	90.7	94.2
livebench_reasoning	82.7	88.1
MCP Atlas	74.2	70.6
MCPAtlas Public (Pass@1)	73.6	67.2
MMLU-Pro	87.5	87.5
nl2repo	38.5	41.3
OmniScience Accuracy	43	50.9
OmniScience Non-Hallucination	12.2	17.4
scicode	50	56.6
simpleqa_verified	57.9	45.3
SWE-bench Multilingual	76.2	71.7
SWE-bench Pro	55.4	59.1
TauBench V3 - Banking	30.1	39.6
Terminal-Bench 2.0	67.9	75.1
Terminal-Bench 2.1	72.1	78.3
Terminal-Bench Hard	46.2	57.6
tool_decathlon	52.8	54.6
toolathlon	51.8	54.6
usamo_2026	60.7	95.2
vectara_answer_rate	97.2	99.9
vectara_avg_summary_length	153.8	81.7
vectara_factual_consistency	91.4	93
vectara_hallucination_rate ↓	8.6	7
τ²-Bench Telecom (AA run)	96.2	87.1
τ³-Bench	25.8	72.9

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.