Which is better, DeepSeek-V4-Pro or GPT-5.5?

Across 56 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 8 and GPT-5.5 leads on 47 — GPT-5.5 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-5.5

DeepSeekvsOpenAI56 shared benchmarks8–47 head-to-head

Benchmark	DeepSeek-V4-Pro	GPT-5.5
AA Agentic Index	63.3	47.4
AA Intelligence	53	56.3
AA-LCR	70	79
AA-Omniscience	-10.6	20.5
Agents' Last Exam	16.5	26.6
aime_2026	96.7	100
apexAgents	24.3	37.7
arena_elo	1457	1482
arena_text_factuality	1450	1482
Artificial Analysis Coding Index	59.4	74.9
browsecomp	83.4	84.4
coding_arena_elo	1582	1457
critpt	13	27.1
cybergym	52.7	81.8
DeepSWE	12.8	70
gdpval	49	49.5
GDPval-AA (Elo)	1554	1769
GDPval-AA v2	1307	1509
GPQA Diamond	90.5	93.6
HLE	48.2	52.2
HLE (with tools)	48.2	52.2
hmmt_feb_2026	95.2	98.5
hmmt_nov_2025	94.4	96.5
IFBench	76.5	75.9
itbenchSre	38.3	45.8
livebench	73.6	80.7
livebench_agentic_coding	42.6	54
livebench_coding	70	82.2
livebench_data_analysis	74.5	81.6
livebench_instruction_following	62.4	70.7
livebench_language	78.1	87.4
livebench_math	90.7	95.9
livebench_reasoning	82.7	89.7
MCP Atlas	74.2	82.8
nl2repo	38.5	50.7
OmniScience Accuracy	43	58
OmniScience Non-Hallucination	12.2	12
Program Bench	47.8	70.8
scicode	50	56.1
simplebench	50.9	69
simpleqa_verified	57.9	63.1
SWE-bench Pro	55.4	58.6
SWE-bench Verified	80.6	80.6
TauBench V3 - Banking	30.1	39
Terminal-Bench 2.0	67.9	82.7
Terminal-Bench 2.1	72.1	84.3
Terminal-Bench Hard	46.2	60.6
tool_decathlon	52.8	55.6
toolathlon	51.8	55.6
usamo_2026	60.7	98.2
vectara_answer_rate	97.2	100
vectara_avg_summary_length	153.8	129.6
vectara_factual_consistency	91.4	90.7
vectara_hallucination_rate ↓	8.6	9.3
τ²-Bench Telecom (AA run)	96.2	93.9
τ³-Bench	25.8	31.3

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.