Which is better, DeepSeek-V4-Pro or Gemini 3 Pro?

Across 44 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 30 and Gemini 3 Pro leads on 14 — 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 Gemini 3 Pro

DeepSeekvsGoogle44 shared benchmarks30–14 head-to-head

Benchmark	DeepSeek-V4-Pro	Gemini 3 Pro
AA Agentic Index	63.3	52
AA Intelligence	53	40.6
AA-LCR	70	73
AA-Omniscience	-10.6	15.3
aime_2026	96.7	91.7
apexAgents	24.3	18.4
arena_elo	1457	1485
arena_text_factuality	1450	1481
Artificial Analysis Coding Index	59.4	46.5
browsecomp	83.4	59.2
coding_arena_elo	1582	1438
critpt	13	9.1
cybergym	52.7	39.9
gdpval	49	34.2
GDPval-AA (Elo)	1554	1195
GPQA Diamond	90.5	91.9
HLE	48.2	45.8
HLE (with tools)	48.2	45.8
hmmt_feb_2026	95.2	86.4
hmmt_nov_2025	94.4	93.3
IFBench	76.5	70.4
imo_answer_bench	89.8	83.3
livebench	73.6	73.4
LiveCodeBench	93.5	92
MCP Atlas	74.2	70.3
MMLU-Pro	87.5	90.1
OmniScience Accuracy	43	55.8
OmniScience Non-Hallucination	12.2	10
scicode	50	56.1
simplebench	50.9	76.4
simpleqa	57.9	72.1
simpleqa_verified	57.9	72.9
SWE-bench Multilingual	76.2	68.7
SWE-bench Pro	55.4	43.3
SWE-bench Verified	80.6	78
Terminal-Bench 2.0	67.9	54.2
Terminal-Bench Hard	46.2	56.9
tool_decathlon	52.8	36.4
toolathlon	51.8	36.4
vectara_answer_rate	97.2	99.4
vectara_avg_summary_length	153.8	101.9
vectara_factual_consistency	91.4	86.4
vectara_hallucination_rate ↓	8.6	13.6
τ²-Bench Telecom (AA run)	96.2	98

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.