Which is better, DeepSeek-V4-Pro or Gemini 3.1 Pro?

Across 82 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 31 and Gemini 3.1 Pro leads on 50 — Gemini 3.1 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 3.1 Pro

DeepSeekvsGoogle82 shared benchmarks31–50 head-to-head

Benchmark	DeepSeek-V4-Pro	Gemini 3.1 Pro
AA Agentic Index	63.3	23
AA Intelligence	53	47.7
AA-LCR	70	79
AA-Omniscience	-10.6	32.9
AgentWorldBench - Android	55.2	61.4
AgentWorldBench - MCP	63.3	59.1
AgentWorldBench - OS	63.7	66.9
AgentWorldBench - Overall	53	54.6
AgentWorldBench - Search	27.6	30.2
AgentWorldBench - SWE	59.4	59.1
AgentWorldBench - Terminal	51.3	52.5
AgentWorldBench - Web	50.3	52.8
aime_2026	96.7	98.3
Apex (Pass@1)	38.3	60.9
Apex Shortlist (Pass@1)	90.2	89.1
apexAgents	24.3	33.5
arena_elo	1457	1486
arena_text_factuality	1450	1471
Artificial Analysis Coding Index	59.4	68.8
browsecomp	83.4	85.9
BrowseComp (w/ Ctx)	83.4	85.9
Chinese SimpleQA (C-SimpleQA)	77.7	85.9
Codeforces	3206	3052
coding_arena_elo	1582	1447
CorpusQA 1M (ACC)	62	53.8
critpt	13	17.7
cybergym	52.7	38.8
DeepSWE	12.8	10
FORTRESS (Adversarial)	36	65.2
FORTRESS (Benign)	98.5	98
gdpval	49	23.3
GDPval-AA (Elo)	1554	1317
GDPval-AA v2	1307	962
Global-MMLU-Lite	89.3	92.7
GPQA Diamond	90.5	94.3
HLE	48.2	51.4
HLE (with tools)	48.2	51.6
hmmt_feb_2026	95.2	94.7
hmmt_nov_2025	94.4	94.8
IFBench	76.5	77.1
imo_answer_bench	89.8	91
itbenchSre	38.3	30.3
livebench	73.6	79.9
livebench_agentic_coding	42.6	44.1
livebench_coding	70	76.5
livebench_data_analysis	74.5	78.5
livebench_instruction_following	62.4	79.1
livebench_language	78.1	85.4
livebench_math	90.7	91
livebench_reasoning	82.7	84
LiveCodeBench	93.5	91.7
MCP Atlas	74.2	78.2
MCPAtlas Public (Pass@1)	73.6	69.2
MMLU-Pro	87.5	91
MRCR 1M (MMR)	83.5	76.3
nl2repo	38.5	33.4
OmniScience Accuracy	43	55.3
OmniScience Non-Hallucination	12.2	50.1
Program Bench	47.8	39.5
scicode	50	59
simplebench	50.9	79.6
simpleqa_verified	57.9	77.3
strongreject	98.6	98
SWE-bench Multilingual	76.2	76.9
SWE-bench Pro	55.4	54.2
SWE-bench Verified	80.6	80.6
SWEBench Pro (Public)	55.4	54.2
Tau 3 Banking	26	16.5
TauBench V3 - Banking	30.1	21.4
Terminal Bench 2.1 (Best Harness)	64	73.8
Terminal-Bench 2.0	67.9	68.5
Terminal-Bench 2.1	72.1	74
Terminal-Bench Hard	46.2	68.5
tool_decathlon	52.8	48.8
toolathlon	51.8	48.8
usamo_2026	60.7	74.4
vectara_answer_rate	97.2	99.4
vectara_avg_summary_length	153.8	107.7
vectara_factual_consistency	91.4	89.6
vectara_hallucination_rate ↓	8.6	10.4
τ²-Bench Telecom (AA run)	96.2	99.3
τ³-Bench	25.8	67.1

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.